fair.c 275.2 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH)
 *
 *  Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
 *
 *  Interactivity improvements by Mike Galbraith
 *  (C) 2007 Mike Galbraith <efault@gmx.de>
 *
 *  Various enhancements by Dmitry Adamushko.
 *  (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com>
 *
 *  Group scheduling enhancements by Srivatsa Vaddagiri
 *  Copyright IBM Corporation, 2007
 *  Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
 *
 *  Scaled math optimizations by Thomas Gleixner
 *  Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de>
19 20
 *
 *  Adaptive scheduling granularity, math enhancements by Peter Zijlstra
21
 *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
22
 */
23
#include "sched.h"
24 25 26

#include <trace/events/sched.h>

27
/*
28
 * Targeted preemption latency for CPU-bound tasks:
29
 *
30
 * NOTE: this latency value is not the same as the concept of
I
Ingo Molnar 已提交
31 32 33
 * 'timeslice length' - timeslices in CFS are of variable length
 * and have no persistent notion like in traditional, time-slice
 * based scheduling concepts.
34
 *
I
Ingo Molnar 已提交
35 36
 * (to see the precise effective timeslice length of your workload,
 *  run vmstat and monitor the context-switches (cs) field)
37 38
 *
 * (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
39
 */
40 41
unsigned int sysctl_sched_latency			= 6000000ULL;
unsigned int normalized_sysctl_sched_latency		= 6000000ULL;
42

43 44 45 46
/*
 * The initial- and re-scaling of tunables is configurable
 *
 * Options are:
47 48 49 50 51 52
 *
 *   SCHED_TUNABLESCALING_NONE - unscaled, always *1
 *   SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
 *   SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
 *
 * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
53
 */
54
enum sched_tunable_scaling sysctl_sched_tunable_scaling = SCHED_TUNABLESCALING_LOG;
55

56
/*
57
 * Minimal preemption granularity for CPU-bound tasks:
58
 *
59
 * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
60
 */
61 62
unsigned int sysctl_sched_min_granularity		= 750000ULL;
unsigned int normalized_sysctl_sched_min_granularity	= 750000ULL;
63 64

/*
65
 * This value is kept at sysctl_sched_latency/sysctl_sched_min_granularity
66
 */
67
static unsigned int sched_nr_latency = 8;
68 69

/*
70
 * After fork, child runs first. If set to 0 (default) then
71
 * parent will (try to) run first.
72
 */
73
unsigned int sysctl_sched_child_runs_first __read_mostly;
74 75 76 77 78 79 80

/*
 * SCHED_OTHER wake-up granularity.
 *
 * This option delays the preemption effects of decoupled workloads
 * and reduces their over-scheduling. Synchronous workloads will still
 * have immediate wakeup/sleep latencies.
81 82
 *
 * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds)
83
 */
84 85
unsigned int sysctl_sched_wakeup_granularity		= 1000000UL;
unsigned int normalized_sysctl_sched_wakeup_granularity	= 1000000UL;
86

87
const_debug unsigned int sysctl_sched_migration_cost	= 500000UL;
88

T
Tim Chen 已提交
89 90
#ifdef CONFIG_SMP
/*
91
 * For asym packing, by default the lower numbered CPU has higher priority.
T
Tim Chen 已提交
92 93 94 95 96 97 98
 */
int __weak arch_asym_cpu_priority(int cpu)
{
	return -cpu;
}
#endif

99 100 101 102 103 104 105 106 107
#ifdef CONFIG_CFS_BANDWIDTH
/*
 * Amount of runtime to allocate from global (tg) to local (per-cfs_rq) pool
 * each time a cfs_rq requests quota.
 *
 * Note: in the case that the slice exceeds the runtime remaining (either due
 * to consumption or the quota being specified to be smaller than the slice)
 * we will always only issue the remaining available time.
 *
108 109 110
 * (default: 5 msec, units: microseconds)
 */
unsigned int sysctl_sched_cfs_bandwidth_slice		= 5000UL;
111 112
#endif

113 114
/*
 * The margin used when comparing utilization with CPU capacity:
115
 * util * margin < capacity * 1024
116 117
 *
 * (default: ~20%)
118
 */
119
unsigned int capacity_margin				= 1280;
120

121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138
static inline void update_load_add(struct load_weight *lw, unsigned long inc)
{
	lw->weight += inc;
	lw->inv_weight = 0;
}

static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
{
	lw->weight -= dec;
	lw->inv_weight = 0;
}

static inline void update_load_set(struct load_weight *lw, unsigned long w)
{
	lw->weight = w;
	lw->inv_weight = 0;
}

139 140 141 142 143 144 145 146 147
/*
 * Increase the granularity value when there are more CPUs,
 * because with more CPUs the 'effective latency' as visible
 * to users decreases. But the relationship is not linear,
 * so pick a second-best guess by going with the log2 of the
 * number of CPUs.
 *
 * This idea comes from the SD scheduler of Con Kolivas:
 */
148
static unsigned int get_update_sysctl_factor(void)
149
{
150
	unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8);
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
	unsigned int factor;

	switch (sysctl_sched_tunable_scaling) {
	case SCHED_TUNABLESCALING_NONE:
		factor = 1;
		break;
	case SCHED_TUNABLESCALING_LINEAR:
		factor = cpus;
		break;
	case SCHED_TUNABLESCALING_LOG:
	default:
		factor = 1 + ilog2(cpus);
		break;
	}

	return factor;
}

static void update_sysctl(void)
{
	unsigned int factor = get_update_sysctl_factor();

#define SET_SYSCTL(name) \
	(sysctl_##name = (factor) * normalized_sysctl_##name)
	SET_SYSCTL(sched_min_granularity);
	SET_SYSCTL(sched_latency);
	SET_SYSCTL(sched_wakeup_granularity);
#undef SET_SYSCTL
}

void sched_init_granularity(void)
{
	update_sysctl();
}

186
#define WMULT_CONST	(~0U)
187 188
#define WMULT_SHIFT	32

189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204
static void __update_inv_weight(struct load_weight *lw)
{
	unsigned long w;

	if (likely(lw->inv_weight))
		return;

	w = scale_load_down(lw->weight);

	if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST))
		lw->inv_weight = 1;
	else if (unlikely(!w))
		lw->inv_weight = WMULT_CONST;
	else
		lw->inv_weight = WMULT_CONST / w;
}
205 206

/*
207 208 209 210
 * delta_exec * weight / lw.weight
 *   OR
 * (delta_exec * (weight * lw->inv_weight)) >> WMULT_SHIFT
 *
211
 * Either weight := NICE_0_LOAD and lw \e sched_prio_to_wmult[], in which case
212 213 214 215 216
 * we're guaranteed shift stays positive because inv_weight is guaranteed to
 * fit 32 bits, and NICE_0_LOAD gives another 10 bits; therefore shift >= 22.
 *
 * Or, weight =< lw.weight (because lw.weight is the runqueue weight), thus
 * weight/lw.weight <= 1, and therefore our shift will also be positive.
217
 */
218
static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight *lw)
219
{
220 221
	u64 fact = scale_load_down(weight);
	int shift = WMULT_SHIFT;
222

223
	__update_inv_weight(lw);
224

225 226 227 228 229
	if (unlikely(fact >> 32)) {
		while (fact >> 32) {
			fact >>= 1;
			shift--;
		}
230 231
	}

232 233
	/* hint to use a 32x32->64 mul */
	fact = (u64)(u32)fact * lw->inv_weight;
234

235 236 237 238
	while (fact >> 32) {
		fact >>= 1;
		shift--;
	}
239

240
	return mul_u64_u32_shr(delta_exec, fact, shift);
241 242 243 244
}


const struct sched_class fair_sched_class;
245

246 247 248 249
/**************************************************************
 * CFS operations on generic schedulable entities:
 */

250
#ifdef CONFIG_FAIR_GROUP_SCHED
251

252
/* cpu runqueue to which this cfs_rq is attached */
253 254
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
255
	return cfs_rq->rq;
256 257
}

258 259
/* An entity is a task if it doesn't "own" a runqueue */
#define entity_is_task(se)	(!se->my_q)
260

261 262
static inline struct task_struct *task_of(struct sched_entity *se)
{
263
	SCHED_WARN_ON(!entity_is_task(se));
264 265 266
	return container_of(se, struct task_struct, se);
}

P
Peter Zijlstra 已提交
267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
/* Walk up scheduling entities hierarchy */
#define for_each_sched_entity(se) \
		for (; se; se = se->parent)

static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
{
	return p->se.cfs_rq;
}

/* runqueue on which this entity is (to be) queued */
static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
{
	return se->cfs_rq;
}

/* runqueue "owned" by this group */
static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
{
	return grp->my_q;
}

288 289 290
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (!cfs_rq->on_list) {
291 292
		struct rq *rq = rq_of(cfs_rq);
		int cpu = cpu_of(rq);
293 294 295
		/*
		 * Ensure we either appear before our parent (if already
		 * enqueued) or force our parent to appear after us when it is
296 297 298 299 300
		 * enqueued. The fact that we always enqueue bottom-up
		 * reduces this to two cases and a special case for the root
		 * cfs_rq. Furthermore, it also means that we will always reset
		 * tmp_alone_branch either when the branch is connected
		 * to a tree or when we reach the beg of the tree
301 302
		 */
		if (cfs_rq->tg->parent &&
303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322
		    cfs_rq->tg->parent->cfs_rq[cpu]->on_list) {
			/*
			 * If parent is already on the list, we add the child
			 * just before. Thanks to circular linked property of
			 * the list, this means to put the child at the tail
			 * of the list that starts by parent.
			 */
			list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
				&(cfs_rq->tg->parent->cfs_rq[cpu]->leaf_cfs_rq_list));
			/*
			 * The branch is now connected to its tree so we can
			 * reset tmp_alone_branch to the beginning of the
			 * list.
			 */
			rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
		} else if (!cfs_rq->tg->parent) {
			/*
			 * cfs rq without parent should be put
			 * at the tail of the list.
			 */
323
			list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343
				&rq->leaf_cfs_rq_list);
			/*
			 * We have reach the beg of a tree so we can reset
			 * tmp_alone_branch to the beginning of the list.
			 */
			rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
		} else {
			/*
			 * The parent has not already been added so we want to
			 * make sure that it will be put after us.
			 * tmp_alone_branch points to the beg of the branch
			 * where we will add parent.
			 */
			list_add_rcu(&cfs_rq->leaf_cfs_rq_list,
				rq->tmp_alone_branch);
			/*
			 * update tmp_alone_branch to points to the new beg
			 * of the branch
			 */
			rq->tmp_alone_branch = &cfs_rq->leaf_cfs_rq_list;
344
		}
345 346 347 348 349 350 351 352 353 354 355 356 357

		cfs_rq->on_list = 1;
	}
}

static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (cfs_rq->on_list) {
		list_del_rcu(&cfs_rq->leaf_cfs_rq_list);
		cfs_rq->on_list = 0;
	}
}

P
Peter Zijlstra 已提交
358
/* Iterate thr' all leaf cfs_rq's on a runqueue */
359 360 361
#define for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos)			\
	list_for_each_entry_safe(cfs_rq, pos, &rq->leaf_cfs_rq_list,	\
				 leaf_cfs_rq_list)
P
Peter Zijlstra 已提交
362 363

/* Do the two (enqueued) entities belong to the same group ? */
P
Peter Zijlstra 已提交
364
static inline struct cfs_rq *
P
Peter Zijlstra 已提交
365 366 367
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	if (se->cfs_rq == pse->cfs_rq)
P
Peter Zijlstra 已提交
368
		return se->cfs_rq;
P
Peter Zijlstra 已提交
369

P
Peter Zijlstra 已提交
370
	return NULL;
P
Peter Zijlstra 已提交
371 372 373 374 375 376 377
}

static inline struct sched_entity *parent_entity(struct sched_entity *se)
{
	return se->parent;
}

378 379 380 381 382 383 384 385 386 387 388 389 390
static void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
	int se_depth, pse_depth;

	/*
	 * preemption test can be made between sibling entities who are in the
	 * same cfs_rq i.e who have a common parent. Walk up the hierarchy of
	 * both tasks until we find their ancestors who are siblings of common
	 * parent.
	 */

	/* First walk up until both entities are at same depth */
P
Peter Zijlstra 已提交
391 392
	se_depth = (*se)->depth;
	pse_depth = (*pse)->depth;
393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409

	while (se_depth > pse_depth) {
		se_depth--;
		*se = parent_entity(*se);
	}

	while (pse_depth > se_depth) {
		pse_depth--;
		*pse = parent_entity(*pse);
	}

	while (!is_same_group(*se, *pse)) {
		*se = parent_entity(*se);
		*pse = parent_entity(*pse);
	}
}

410 411 412 413 414 415
#else	/* !CONFIG_FAIR_GROUP_SCHED */

static inline struct task_struct *task_of(struct sched_entity *se)
{
	return container_of(se, struct task_struct, se);
}
416

417 418 419
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
	return container_of(cfs_rq, struct rq, cfs);
420 421 422 423
}

#define entity_is_task(se)	1

P
Peter Zijlstra 已提交
424 425
#define for_each_sched_entity(se) \
		for (; se; se = NULL)
426

P
Peter Zijlstra 已提交
427
static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
428
{
P
Peter Zijlstra 已提交
429
	return &task_rq(p)->cfs;
430 431
}

P
Peter Zijlstra 已提交
432 433 434 435 436 437 438 439 440 441 442 443 444 445
static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
{
	struct task_struct *p = task_of(se);
	struct rq *rq = task_rq(p);

	return &rq->cfs;
}

/* runqueue "owned" by this group */
static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
{
	return NULL;
}

446 447 448 449 450 451 452 453
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
}

static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
}

454 455
#define for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos)	\
		for (cfs_rq = &rq->cfs, pos = NULL; cfs_rq; cfs_rq = pos)
P
Peter Zijlstra 已提交
456 457 458 459 460 461

static inline struct sched_entity *parent_entity(struct sched_entity *se)
{
	return NULL;
}

462 463 464 465 466
static inline void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
}

P
Peter Zijlstra 已提交
467 468
#endif	/* CONFIG_FAIR_GROUP_SCHED */

469
static __always_inline
470
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec);
471 472 473 474 475

/**************************************************************
 * Scheduling class tree data structure manipulation methods:
 */

476
static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
477
{
478
	s64 delta = (s64)(vruntime - max_vruntime);
479
	if (delta > 0)
480
		max_vruntime = vruntime;
481

482
	return max_vruntime;
483 484
}

485
static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
P
Peter Zijlstra 已提交
486 487 488 489 490 491 492 493
{
	s64 delta = (s64)(vruntime - min_vruntime);
	if (delta < 0)
		min_vruntime = vruntime;

	return min_vruntime;
}

494 495 496 497 498 499
static inline int entity_before(struct sched_entity *a,
				struct sched_entity *b)
{
	return (s64)(a->vruntime - b->vruntime) < 0;
}

500 501
static void update_min_vruntime(struct cfs_rq *cfs_rq)
{
502
	struct sched_entity *curr = cfs_rq->curr;
503
	struct rb_node *leftmost = rb_first_cached(&cfs_rq->tasks_timeline);
504

505 506
	u64 vruntime = cfs_rq->min_vruntime;

507 508 509 510 511 512
	if (curr) {
		if (curr->on_rq)
			vruntime = curr->vruntime;
		else
			curr = NULL;
	}
513

514 515 516
	if (leftmost) { /* non-empty tree */
		struct sched_entity *se;
		se = rb_entry(leftmost, struct sched_entity, run_node);
517

518
		if (!curr)
519 520 521 522 523
			vruntime = se->vruntime;
		else
			vruntime = min_vruntime(vruntime, se->vruntime);
	}

524
	/* ensure we never gain time by being placed backwards. */
525
	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
526 527 528 529
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
530 531
}

532 533 534
/*
 * Enqueue an entity into the rb-tree:
 */
535
static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
536
{
537
	struct rb_node **link = &cfs_rq->tasks_timeline.rb_root.rb_node;
538 539
	struct rb_node *parent = NULL;
	struct sched_entity *entry;
540
	bool leftmost = true;
541 542 543 544 545 546 547 548 549 550 551

	/*
	 * Find the right place in the rbtree:
	 */
	while (*link) {
		parent = *link;
		entry = rb_entry(parent, struct sched_entity, run_node);
		/*
		 * We dont care about collisions. Nodes with
		 * the same key stay together.
		 */
552
		if (entity_before(se, entry)) {
553 554 555
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
556
			leftmost = false;
557 558 559 560
		}
	}

	rb_link_node(&se->run_node, parent, link);
561 562
	rb_insert_color_cached(&se->run_node,
			       &cfs_rq->tasks_timeline, leftmost);
563 564
}

565
static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
566
{
567
	rb_erase_cached(&se->run_node, &cfs_rq->tasks_timeline);
568 569
}

570
struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
571
{
572
	struct rb_node *left = rb_first_cached(&cfs_rq->tasks_timeline);
573 574 575 576 577

	if (!left)
		return NULL;

	return rb_entry(left, struct sched_entity, run_node);
578 579
}

580 581 582 583 584 585 586 587 588 589 590
static struct sched_entity *__pick_next_entity(struct sched_entity *se)
{
	struct rb_node *next = rb_next(&se->run_node);

	if (!next)
		return NULL;

	return rb_entry(next, struct sched_entity, run_node);
}

#ifdef CONFIG_SCHED_DEBUG
591
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
592
{
593
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline.rb_root);
594

595 596
	if (!last)
		return NULL;
597 598

	return rb_entry(last, struct sched_entity, run_node);
599 600
}

601 602 603 604
/**************************************************************
 * Scheduling class statistics methods:
 */

605
int sched_proc_update_handler(struct ctl_table *table, int write,
606
		void __user *buffer, size_t *lenp,
607 608
		loff_t *ppos)
{
609
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
610
	unsigned int factor = get_update_sysctl_factor();
611 612 613 614 615 616 617

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

618 619 620 621 622 623 624
#define WRT_SYSCTL(name) \
	(normalized_sysctl_##name = sysctl_##name / (factor))
	WRT_SYSCTL(sched_min_granularity);
	WRT_SYSCTL(sched_latency);
	WRT_SYSCTL(sched_wakeup_granularity);
#undef WRT_SYSCTL

625 626 627
	return 0;
}
#endif
628

629
/*
630
 * delta /= w
631
 */
632
static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se)
633
{
634
	if (unlikely(se->load.weight != NICE_0_LOAD))
635
		delta = __calc_delta(delta, NICE_0_LOAD, &se->load);
636 637 638 639

	return delta;
}

640 641 642
/*
 * The idea is to set a period in which each task runs once.
 *
643
 * When there are too many tasks (sched_nr_latency) we have to stretch
644 645 646 647
 * this period because otherwise the slices get too small.
 *
 * p = (nr <= nl) ? l : l*nr/nl
 */
648 649
static u64 __sched_period(unsigned long nr_running)
{
650 651 652 653
	if (unlikely(nr_running > sched_nr_latency))
		return nr_running * sysctl_sched_min_granularity;
	else
		return sysctl_sched_latency;
654 655
}

656 657 658 659
/*
 * We calculate the wall-time slice from the period by taking a part
 * proportional to the weight.
 *
660
 * s = p*P[w/rw]
661
 */
P
Peter Zijlstra 已提交
662
static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
663
{
M
Mike Galbraith 已提交
664
	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
665

M
Mike Galbraith 已提交
666
	for_each_sched_entity(se) {
L
Lin Ming 已提交
667
		struct load_weight *load;
668
		struct load_weight lw;
L
Lin Ming 已提交
669 670 671

		cfs_rq = cfs_rq_of(se);
		load = &cfs_rq->load;
672

M
Mike Galbraith 已提交
673
		if (unlikely(!se->on_rq)) {
674
			lw = cfs_rq->load;
M
Mike Galbraith 已提交
675 676 677 678

			update_load_add(&lw, se->load.weight);
			load = &lw;
		}
679
		slice = __calc_delta(slice, se->load.weight, load);
M
Mike Galbraith 已提交
680 681
	}
	return slice;
682 683
}

684
/*
A
Andrei Epure 已提交
685
 * We calculate the vruntime slice of a to-be-inserted task.
686
 *
687
 * vs = s/w
688
 */
689
static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
P
Peter Zijlstra 已提交
690
{
691
	return calc_delta_fair(sched_slice(cfs_rq, se), se);
692 693
}

694
#ifdef CONFIG_SMP
695 696 697

#include "sched-pelt.h"

698
static int select_idle_sibling(struct task_struct *p, int prev_cpu, int cpu);
699 700
static unsigned long task_h_load(struct task_struct *p);

701 702
/* Give new sched_entity start runnable values to heavy its load in infant time */
void init_entity_runnable_average(struct sched_entity *se)
703
{
704
	struct sched_avg *sa = &se->avg;
705

706 707
	memset(sa, 0, sizeof(*sa));

708 709 710 711 712 713 714
	/*
	 * Tasks are intialized with full load to be seen as heavy tasks until
	 * they get a chance to stabilize to their real load level.
	 * Group entities are intialized with zero load to reflect the fact that
	 * nothing has been attached to the task group yet.
	 */
	if (entity_is_task(se))
715 716
		sa->runnable_load_avg = sa->load_avg = scale_load_down(se->load.weight);

717 718
	se->runnable_weight = se->load.weight;

719
	/* when this task enqueue'ed, it will contribute to its cfs_rq's load_avg */
720
}
721

722
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);
723
static void attach_entity_cfs_rq(struct sched_entity *se);
724

725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753
/*
 * With new tasks being created, their initial util_avgs are extrapolated
 * based on the cfs_rq's current util_avg:
 *
 *   util_avg = cfs_rq->util_avg / (cfs_rq->load_avg + 1) * se.load.weight
 *
 * However, in many cases, the above util_avg does not give a desired
 * value. Moreover, the sum of the util_avgs may be divergent, such
 * as when the series is a harmonic series.
 *
 * To solve this problem, we also cap the util_avg of successive tasks to
 * only 1/2 of the left utilization budget:
 *
 *   util_avg_cap = (1024 - cfs_rq->avg.util_avg) / 2^n
 *
 * where n denotes the nth task.
 *
 * For example, a simplest series from the beginning would be like:
 *
 *  task  util_avg: 512, 256, 128,  64,  32,   16,    8, ...
 * cfs_rq util_avg: 512, 768, 896, 960, 992, 1008, 1016, ...
 *
 * Finally, that extrapolated util_avg is clamped to the cap (util_avg_cap)
 * if util_avg > util_avg_cap.
 */
void post_init_entity_util_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	struct sched_avg *sa = &se->avg;
754
	long cap = (long)(SCHED_CAPACITY_SCALE - cfs_rq->avg.util_avg) / 2;
755 756 757 758 759 760 761 762 763 764 765 766

	if (cap > 0) {
		if (cfs_rq->avg.util_avg != 0) {
			sa->util_avg  = cfs_rq->avg.util_avg * se->load.weight;
			sa->util_avg /= (cfs_rq->avg.load_avg + 1);

			if (sa->util_avg > cap)
				sa->util_avg = cap;
		} else {
			sa->util_avg = cap;
		}
	}
767 768 769 770 771 772 773

	if (entity_is_task(se)) {
		struct task_struct *p = task_of(se);
		if (p->sched_class != &fair_sched_class) {
			/*
			 * For !fair tasks do:
			 *
774
			update_cfs_rq_load_avg(now, cfs_rq);
775
			attach_entity_load_avg(cfs_rq, se, 0);
776 777 778 779 780
			switched_from_fair(rq, p);
			 *
			 * such that the next switched_to_fair() has the
			 * expected state.
			 */
781
			se->avg.last_update_time = cfs_rq_clock_task(cfs_rq);
782 783 784 785
			return;
		}
	}

786
	attach_entity_cfs_rq(se);
787 788
}

789
#else /* !CONFIG_SMP */
790
void init_entity_runnable_average(struct sched_entity *se)
791 792
{
}
793 794 795
void post_init_entity_util_avg(struct sched_entity *se)
{
}
796 797 798
static void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
{
}
799
#endif /* CONFIG_SMP */
800

801
/*
802
 * Update the current task's runtime statistics.
803
 */
804
static void update_curr(struct cfs_rq *cfs_rq)
805
{
806
	struct sched_entity *curr = cfs_rq->curr;
807
	u64 now = rq_clock_task(rq_of(cfs_rq));
808
	u64 delta_exec;
809 810 811 812

	if (unlikely(!curr))
		return;

813 814
	delta_exec = now - curr->exec_start;
	if (unlikely((s64)delta_exec <= 0))
P
Peter Zijlstra 已提交
815
		return;
816

I
Ingo Molnar 已提交
817
	curr->exec_start = now;
818

819 820 821 822
	schedstat_set(curr->statistics.exec_max,
		      max(delta_exec, curr->statistics.exec_max));

	curr->sum_exec_runtime += delta_exec;
823
	schedstat_add(cfs_rq->exec_clock, delta_exec);
824 825 826 827

	curr->vruntime += calc_delta_fair(delta_exec, curr);
	update_min_vruntime(cfs_rq);

828 829 830
	if (entity_is_task(curr)) {
		struct task_struct *curtask = task_of(curr);

831
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
832
		cgroup_account_cputime(curtask, delta_exec);
833
		account_group_exec_runtime(curtask, delta_exec);
834
	}
835 836

	account_cfs_rq_runtime(cfs_rq, delta_exec);
837 838
}

839 840 841 842 843
static void update_curr_fair(struct rq *rq)
{
	update_curr(cfs_rq_of(&rq->curr->se));
}

844
static inline void
845
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
846
{
847 848 849 850 851 852 853
	u64 wait_start, prev_wait_start;

	if (!schedstat_enabled())
		return;

	wait_start = rq_clock(rq_of(cfs_rq));
	prev_wait_start = schedstat_val(se->statistics.wait_start);
854 855

	if (entity_is_task(se) && task_on_rq_migrating(task_of(se)) &&
856 857
	    likely(wait_start > prev_wait_start))
		wait_start -= prev_wait_start;
858

859
	__schedstat_set(se->statistics.wait_start, wait_start);
860 861
}

862
static inline void
863 864 865
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *p;
866 867
	u64 delta;

868 869 870 871
	if (!schedstat_enabled())
		return;

	delta = rq_clock(rq_of(cfs_rq)) - schedstat_val(se->statistics.wait_start);
872 873 874 875 876 877 878 879 880

	if (entity_is_task(se)) {
		p = task_of(se);
		if (task_on_rq_migrating(p)) {
			/*
			 * Preserve migrating task's wait time so wait_start
			 * time stamp can be adjusted to accumulate wait time
			 * prior to migration.
			 */
881
			__schedstat_set(se->statistics.wait_start, delta);
882 883 884 885 886
			return;
		}
		trace_sched_stat_wait(p, delta);
	}

887
	__schedstat_set(se->statistics.wait_max,
888
		      max(schedstat_val(se->statistics.wait_max), delta));
889 890 891
	__schedstat_inc(se->statistics.wait_count);
	__schedstat_add(se->statistics.wait_sum, delta);
	__schedstat_set(se->statistics.wait_start, 0);
892 893
}

894
static inline void
895 896 897
update_stats_enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *tsk = NULL;
898 899 900 901 902 903 904
	u64 sleep_start, block_start;

	if (!schedstat_enabled())
		return;

	sleep_start = schedstat_val(se->statistics.sleep_start);
	block_start = schedstat_val(se->statistics.block_start);
905 906 907 908

	if (entity_is_task(se))
		tsk = task_of(se);

909 910
	if (sleep_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - sleep_start;
911 912 913 914

		if ((s64)delta < 0)
			delta = 0;

915
		if (unlikely(delta > schedstat_val(se->statistics.sleep_max)))
916
			__schedstat_set(se->statistics.sleep_max, delta);
917

918 919
		__schedstat_set(se->statistics.sleep_start, 0);
		__schedstat_add(se->statistics.sum_sleep_runtime, delta);
920 921 922 923 924 925

		if (tsk) {
			account_scheduler_latency(tsk, delta >> 10, 1);
			trace_sched_stat_sleep(tsk, delta);
		}
	}
926 927
	if (block_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - block_start;
928 929 930 931

		if ((s64)delta < 0)
			delta = 0;

932
		if (unlikely(delta > schedstat_val(se->statistics.block_max)))
933
			__schedstat_set(se->statistics.block_max, delta);
934

935 936
		__schedstat_set(se->statistics.block_start, 0);
		__schedstat_add(se->statistics.sum_sleep_runtime, delta);
937 938 939

		if (tsk) {
			if (tsk->in_iowait) {
940 941
				__schedstat_add(se->statistics.iowait_sum, delta);
				__schedstat_inc(se->statistics.iowait_count);
942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959
				trace_sched_stat_iowait(tsk, delta);
			}

			trace_sched_stat_blocked(tsk, delta);

			/*
			 * Blocking time is in units of nanosecs, so shift by
			 * 20 to get a milliseconds-range estimation of the
			 * amount of time that the task spent sleeping:
			 */
			if (unlikely(prof_on == SLEEP_PROFILING)) {
				profile_hits(SLEEP_PROFILING,
						(void *)get_wchan(tsk),
						delta >> 20);
			}
			account_scheduler_latency(tsk, delta >> 10, 0);
		}
	}
960 961
}

962 963 964
/*
 * Task is being enqueued - update stats:
 */
965
static inline void
966
update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
967
{
968 969 970
	if (!schedstat_enabled())
		return;

971 972 973 974
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
975
	if (se != cfs_rq->curr)
976
		update_stats_wait_start(cfs_rq, se);
977 978 979

	if (flags & ENQUEUE_WAKEUP)
		update_stats_enqueue_sleeper(cfs_rq, se);
980 981 982
}

static inline void
983
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
984
{
985 986 987 988

	if (!schedstat_enabled())
		return;

989 990 991 992
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
993
	if (se != cfs_rq->curr)
994
		update_stats_wait_end(cfs_rq, se);
995

996 997
	if ((flags & DEQUEUE_SLEEP) && entity_is_task(se)) {
		struct task_struct *tsk = task_of(se);
998

999
		if (tsk->state & TASK_INTERRUPTIBLE)
1000
			__schedstat_set(se->statistics.sleep_start,
1001 1002
				      rq_clock(rq_of(cfs_rq)));
		if (tsk->state & TASK_UNINTERRUPTIBLE)
1003
			__schedstat_set(se->statistics.block_start,
1004
				      rq_clock(rq_of(cfs_rq)));
1005 1006 1007
	}
}

1008 1009 1010 1011
/*
 * We are picking a new current task - update its stats:
 */
static inline void
1012
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
1013 1014 1015 1016
{
	/*
	 * We are starting a new run period:
	 */
1017
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
1018 1019 1020 1021 1022 1023
}

/**************************************************
 * Scheduling class queueing methods:
 */

1024 1025
#ifdef CONFIG_NUMA_BALANCING
/*
1026 1027 1028
 * Approximate time to scan a full NUMA task in ms. The task scan period is
 * calculated based on the tasks virtual memory size and
 * numa_balancing_scan_size.
1029
 */
1030 1031
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
1032 1033 1034

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
1035

1036 1037 1038
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061
struct numa_group {
	atomic_t refcount;

	spinlock_t lock; /* nr_tasks, tasks */
	int nr_tasks;
	pid_t gid;
	int active_nodes;

	struct rcu_head rcu;
	unsigned long total_faults;
	unsigned long max_faults_cpu;
	/*
	 * Faults_cpu is used to decide whether memory should move
	 * towards the CPU. As a consequence, these stats are weighted
	 * more by CPU use than by memory faults.
	 */
	unsigned long *faults_cpu;
	unsigned long faults[0];
};

static inline unsigned long group_faults_priv(struct numa_group *ng);
static inline unsigned long group_faults_shared(struct numa_group *ng);

1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085
static unsigned int task_nr_scan_windows(struct task_struct *p)
{
	unsigned long rss = 0;
	unsigned long nr_scan_pages;

	/*
	 * Calculations based on RSS as non-present and empty pages are skipped
	 * by the PTE scanner and NUMA hinting faults should be trapped based
	 * on resident pages
	 */
	nr_scan_pages = sysctl_numa_balancing_scan_size << (20 - PAGE_SHIFT);
	rss = get_mm_rss(p->mm);
	if (!rss)
		rss = nr_scan_pages;

	rss = round_up(rss, nr_scan_pages);
	return rss / nr_scan_pages;
}

/* For sanitys sake, never scan more PTEs than MAX_SCAN_WINDOW MB/sec. */
#define MAX_SCAN_WINDOW 2560

static unsigned int task_scan_min(struct task_struct *p)
{
1086
	unsigned int scan_size = READ_ONCE(sysctl_numa_balancing_scan_size);
1087 1088 1089
	unsigned int scan, floor;
	unsigned int windows = 1;

1090 1091
	if (scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / scan_size;
1092 1093 1094 1095 1096 1097
	floor = 1000 / windows;

	scan = sysctl_numa_balancing_scan_period_min / task_nr_scan_windows(p);
	return max_t(unsigned int, floor, scan);
}

1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116
static unsigned int task_scan_start(struct task_struct *p)
{
	unsigned long smin = task_scan_min(p);
	unsigned long period = smin;

	/* Scale the maximum scan period with the amount of shared memory. */
	if (p->numa_group) {
		struct numa_group *ng = p->numa_group;
		unsigned long shared = group_faults_shared(ng);
		unsigned long private = group_faults_priv(ng);

		period *= atomic_read(&ng->refcount);
		period *= shared + 1;
		period /= private + shared + 1;
	}

	return max(smin, period);
}

1117 1118
static unsigned int task_scan_max(struct task_struct *p)
{
1119 1120
	unsigned long smin = task_scan_min(p);
	unsigned long smax;
1121 1122 1123

	/* Watch for min being lower than max due to floor calculations */
	smax = sysctl_numa_balancing_scan_period_max / task_nr_scan_windows(p);
1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138

	/* Scale the maximum scan period with the amount of shared memory. */
	if (p->numa_group) {
		struct numa_group *ng = p->numa_group;
		unsigned long shared = group_faults_shared(ng);
		unsigned long private = group_faults_priv(ng);
		unsigned long period = smax;

		period *= atomic_read(&ng->refcount);
		period *= shared + 1;
		period /= private + shared + 1;

		smax = max(smax, period);
	}

1139 1140 1141
	return max(smin, smax);
}

1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153
static void account_numa_enqueue(struct rq *rq, struct task_struct *p)
{
	rq->nr_numa_running += (p->numa_preferred_nid != -1);
	rq->nr_preferred_running += (p->numa_preferred_nid == task_node(p));
}

static void account_numa_dequeue(struct rq *rq, struct task_struct *p)
{
	rq->nr_numa_running -= (p->numa_preferred_nid != -1);
	rq->nr_preferred_running -= (p->numa_preferred_nid == task_node(p));
}

1154 1155 1156 1157 1158 1159 1160 1161 1162
/* Shared or private faults. */
#define NR_NUMA_HINT_FAULT_TYPES 2

/* Memory and CPU locality */
#define NR_NUMA_HINT_FAULT_STATS (NR_NUMA_HINT_FAULT_TYPES * 2)

/* Averaged statistics, and temporary buffers. */
#define NR_NUMA_HINT_FAULT_BUCKETS (NR_NUMA_HINT_FAULT_STATS * 2)

1163 1164 1165 1166 1167
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

1168
/*
1169
 * The averaged statistics, shared & private, memory & CPU,
1170 1171 1172 1173 1174
 * occupy the first half of the array. The second half of the
 * array is for current counters, which are averaged into the
 * first set by task_numa_placement.
 */
static inline int task_faults_idx(enum numa_faults_stats s, int nid, int priv)
1175
{
1176
	return NR_NUMA_HINT_FAULT_TYPES * (s * nr_node_ids + nid) + priv;
1177 1178 1179 1180
}

static inline unsigned long task_faults(struct task_struct *p, int nid)
{
1181
	if (!p->numa_faults)
1182 1183
		return 0;

1184 1185
	return p->numa_faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_faults[task_faults_idx(NUMA_MEM, nid, 1)];
1186 1187
}

1188 1189 1190 1191 1192
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

1193 1194
	return p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 1)];
1195 1196
}

1197 1198
static inline unsigned long group_faults_cpu(struct numa_group *group, int nid)
{
1199 1200
	return group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 0)] +
		group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 1)];
1201 1202
}

1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226
static inline unsigned long group_faults_priv(struct numa_group *ng)
{
	unsigned long faults = 0;
	int node;

	for_each_online_node(node) {
		faults += ng->faults[task_faults_idx(NUMA_MEM, node, 1)];
	}

	return faults;
}

static inline unsigned long group_faults_shared(struct numa_group *ng)
{
	unsigned long faults = 0;
	int node;

	for_each_online_node(node) {
		faults += ng->faults[task_faults_idx(NUMA_MEM, node, 0)];
	}

	return faults;
}

1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
/*
 * A node triggering more than 1/3 as many NUMA faults as the maximum is
 * considered part of a numa group's pseudo-interleaving set. Migrations
 * between these nodes are slowed down, to allow things to settle down.
 */
#define ACTIVE_NODE_FRACTION 3

static bool numa_is_active_node(int nid, struct numa_group *ng)
{
	return group_faults_cpu(ng, nid) * ACTIVE_NODE_FRACTION > ng->max_faults_cpu;
}

1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303
/* Handle placement on systems where not all nodes are directly connected. */
static unsigned long score_nearby_nodes(struct task_struct *p, int nid,
					int maxdist, bool task)
{
	unsigned long score = 0;
	int node;

	/*
	 * All nodes are directly connected, and the same distance
	 * from each other. No need for fancy placement algorithms.
	 */
	if (sched_numa_topology_type == NUMA_DIRECT)
		return 0;

	/*
	 * This code is called for each node, introducing N^2 complexity,
	 * which should be ok given the number of nodes rarely exceeds 8.
	 */
	for_each_online_node(node) {
		unsigned long faults;
		int dist = node_distance(nid, node);

		/*
		 * The furthest away nodes in the system are not interesting
		 * for placement; nid was already counted.
		 */
		if (dist == sched_max_numa_distance || node == nid)
			continue;

		/*
		 * On systems with a backplane NUMA topology, compare groups
		 * of nodes, and move tasks towards the group with the most
		 * memory accesses. When comparing two nodes at distance
		 * "hoplimit", only nodes closer by than "hoplimit" are part
		 * of each group. Skip other nodes.
		 */
		if (sched_numa_topology_type == NUMA_BACKPLANE &&
					dist > maxdist)
			continue;

		/* Add up the faults from nearby nodes. */
		if (task)
			faults = task_faults(p, node);
		else
			faults = group_faults(p, node);

		/*
		 * On systems with a glueless mesh NUMA topology, there are
		 * no fixed "groups of nodes". Instead, nodes that are not
		 * directly connected bounce traffic through intermediate
		 * nodes; a numa_group can occupy any set of nodes.
		 * The further away a node is, the less the faults count.
		 * This seems to result in good task placement.
		 */
		if (sched_numa_topology_type == NUMA_GLUELESS_MESH) {
			faults *= (sched_max_numa_distance - dist);
			faults /= (sched_max_numa_distance - LOCAL_DISTANCE);
		}

		score += faults;
	}

	return score;
}

1304 1305 1306 1307 1308 1309
/*
 * These return the fraction of accesses done by a particular task, or
 * task group, on a particular numa node.  The group weight is given a
 * larger multiplier, in order to group tasks together that are almost
 * evenly spread out between numa nodes.
 */
1310 1311
static inline unsigned long task_weight(struct task_struct *p, int nid,
					int dist)
1312
{
1313
	unsigned long faults, total_faults;
1314

1315
	if (!p->numa_faults)
1316 1317 1318 1319 1320 1321 1322
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

1323
	faults = task_faults(p, nid);
1324 1325
	faults += score_nearby_nodes(p, nid, dist, true);

1326
	return 1000 * faults / total_faults;
1327 1328
}

1329 1330
static inline unsigned long group_weight(struct task_struct *p, int nid,
					 int dist)
1331
{
1332 1333 1334 1335 1336 1337 1338 1339
	unsigned long faults, total_faults;

	if (!p->numa_group)
		return 0;

	total_faults = p->numa_group->total_faults;

	if (!total_faults)
1340 1341
		return 0;

1342
	faults = group_faults(p, nid);
1343 1344
	faults += score_nearby_nodes(p, nid, dist, false);

1345
	return 1000 * faults / total_faults;
1346 1347
}

1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387
bool should_numa_migrate_memory(struct task_struct *p, struct page * page,
				int src_nid, int dst_cpu)
{
	struct numa_group *ng = p->numa_group;
	int dst_nid = cpu_to_node(dst_cpu);
	int last_cpupid, this_cpupid;

	this_cpupid = cpu_pid_to_cpupid(dst_cpu, current->pid);

	/*
	 * Multi-stage node selection is used in conjunction with a periodic
	 * migration fault to build a temporal task<->page relation. By using
	 * a two-stage filter we remove short/unlikely relations.
	 *
	 * Using P(p) ~ n_p / n_t as per frequentist probability, we can equate
	 * a task's usage of a particular page (n_p) per total usage of this
	 * page (n_t) (in a given time-span) to a probability.
	 *
	 * Our periodic faults will sample this probability and getting the
	 * same result twice in a row, given these samples are fully
	 * independent, is then given by P(n)^2, provided our sample period
	 * is sufficiently short compared to the usage pattern.
	 *
	 * This quadric squishes small probabilities, making it less likely we
	 * act on an unlikely task<->page relation.
	 */
	last_cpupid = page_cpupid_xchg_last(page, this_cpupid);
	if (!cpupid_pid_unset(last_cpupid) &&
				cpupid_to_nid(last_cpupid) != dst_nid)
		return false;

	/* Always allow migrate on private faults */
	if (cpupid_match_pid(p, last_cpupid))
		return true;

	/* A shared fault, but p->numa_group has not been set up yet. */
	if (!ng)
		return true;

	/*
1388 1389
	 * Destination node is much more heavily used than the source
	 * node? Allow migration.
1390
	 */
1391 1392
	if (group_faults_cpu(ng, dst_nid) > group_faults_cpu(ng, src_nid) *
					ACTIVE_NODE_FRACTION)
1393 1394 1395
		return true;

	/*
1396 1397 1398 1399 1400 1401
	 * Distribute memory according to CPU & memory use on each node,
	 * with 3/4 hysteresis to avoid unnecessary memory migrations:
	 *
	 * faults_cpu(dst)   3   faults_cpu(src)
	 * --------------- * - > ---------------
	 * faults_mem(dst)   4   faults_mem(src)
1402
	 */
1403 1404
	return group_faults_cpu(ng, dst_nid) * group_faults(p, src_nid) * 3 >
	       group_faults_cpu(ng, src_nid) * group_faults(p, dst_nid) * 4;
1405 1406
}

1407
static unsigned long weighted_cpuload(struct rq *rq);
1408 1409
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1410
static unsigned long capacity_of(int cpu);
1411

1412
/* Cached statistics for all CPUs within a node */
1413
struct numa_stats {
1414
	unsigned long nr_running;
1415
	unsigned long load;
1416 1417

	/* Total compute capacity of CPUs on a node */
1418
	unsigned long compute_capacity;
1419 1420

	/* Approximate capacity in terms of runnable tasks on a node */
1421
	unsigned long task_capacity;
1422
	int has_free_capacity;
1423
};
1424

1425 1426 1427 1428 1429
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1430 1431
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1432 1433 1434 1435 1436 1437

	memset(ns, 0, sizeof(*ns));
	for_each_cpu(cpu, cpumask_of_node(nid)) {
		struct rq *rq = cpu_rq(cpu);

		ns->nr_running += rq->nr_running;
1438
		ns->load += weighted_cpuload(rq);
1439
		ns->compute_capacity += capacity_of(cpu);
1440 1441

		cpus++;
1442 1443
	}

1444 1445 1446 1447 1448
	/*
	 * If we raced with hotplug and there are no CPUs left in our mask
	 * the @ns structure is NULL'ed and task_numa_compare() will
	 * not find this node attractive.
	 *
1449 1450
	 * We'll either bail at !has_free_capacity, or we'll detect a huge
	 * imbalance and bail there.
1451 1452 1453 1454
	 */
	if (!cpus)
		return;

1455 1456 1457 1458 1459 1460
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_power < 2 */
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, ns->compute_capacity);
	capacity = cpus / smt; /* cores */

	ns->task_capacity = min_t(unsigned, capacity,
		DIV_ROUND_CLOSEST(ns->compute_capacity, SCHED_CAPACITY_SCALE));
1461
	ns->has_free_capacity = (ns->nr_running < ns->task_capacity);
1462 1463
}

1464 1465
struct task_numa_env {
	struct task_struct *p;
1466

1467 1468
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1469

1470
	struct numa_stats src_stats, dst_stats;
1471

1472
	int imbalance_pct;
1473
	int dist;
1474 1475 1476

	struct task_struct *best_task;
	long best_imp;
1477 1478 1479
	int best_cpu;
};

1480 1481 1482 1483 1484
static void task_numa_assign(struct task_numa_env *env,
			     struct task_struct *p, long imp)
{
	if (env->best_task)
		put_task_struct(env->best_task);
1485 1486
	if (p)
		get_task_struct(p);
1487 1488 1489 1490 1491 1492

	env->best_task = p;
	env->best_imp = imp;
	env->best_cpu = env->dst_cpu;
}

1493
static bool load_too_imbalanced(long src_load, long dst_load,
1494 1495
				struct task_numa_env *env)
{
1496 1497
	long imb, old_imb;
	long orig_src_load, orig_dst_load;
1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508
	long src_capacity, dst_capacity;

	/*
	 * The load is corrected for the CPU capacity available on each node.
	 *
	 * src_load        dst_load
	 * ------------ vs ---------
	 * src_capacity    dst_capacity
	 */
	src_capacity = env->src_stats.compute_capacity;
	dst_capacity = env->dst_stats.compute_capacity;
1509 1510

	/* We care about the slope of the imbalance, not the direction. */
1511 1512
	if (dst_load < src_load)
		swap(dst_load, src_load);
1513 1514

	/* Is the difference below the threshold? */
1515 1516
	imb = dst_load * src_capacity * 100 -
	      src_load * dst_capacity * env->imbalance_pct;
1517 1518 1519 1520 1521
	if (imb <= 0)
		return false;

	/*
	 * The imbalance is above the allowed threshold.
1522
	 * Compare it with the old imbalance.
1523
	 */
1524
	orig_src_load = env->src_stats.load;
1525
	orig_dst_load = env->dst_stats.load;
1526

1527 1528
	if (orig_dst_load < orig_src_load)
		swap(orig_dst_load, orig_src_load);
1529

1530 1531 1532 1533 1534
	old_imb = orig_dst_load * src_capacity * 100 -
		  orig_src_load * dst_capacity * env->imbalance_pct;

	/* Would this change make things worse? */
	return (imb > old_imb);
1535 1536
}

1537 1538 1539 1540 1541 1542
/*
 * This checks if the overall compute and NUMA accesses of the system would
 * be improved if the source tasks was migrated to the target dst_cpu taking
 * into account that it might be best if task running on the dst_cpu should
 * be exchanged with the source task
 */
1543 1544
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1545 1546 1547 1548
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1549
	long src_load, dst_load;
1550
	long load;
1551
	long imp = env->p->numa_group ? groupimp : taskimp;
1552
	long moveimp = imp;
1553
	int dist = env->dist;
1554 1555

	rcu_read_lock();
1556 1557
	cur = task_rcu_dereference(&dst_rq->curr);
	if (cur && ((cur->flags & PF_EXITING) || is_idle_task(cur)))
1558 1559
		cur = NULL;

1560 1561 1562 1563 1564 1565 1566
	/*
	 * Because we have preemption enabled we can get migrated around and
	 * end try selecting ourselves (current == env->p) as a swap candidate.
	 */
	if (cur == env->p)
		goto unlock;

1567 1568 1569 1570 1571 1572 1573 1574
	/*
	 * "imp" is the fault differential for the source task between the
	 * source and destination node. Calculate the total differential for
	 * the source task and potential destination task. The more negative
	 * the value is, the more rmeote accesses that would be expected to
	 * be incurred if the tasks were swapped.
	 */
	if (cur) {
1575
		/* Skip this swap candidate if cannot move to the source CPU: */
1576
		if (!cpumask_test_cpu(env->src_cpu, &cur->cpus_allowed))
1577 1578
			goto unlock;

1579 1580
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1581
		 * in any group then look only at task weights.
1582
		 */
1583
		if (cur->numa_group == env->p->numa_group) {
1584 1585
			imp = taskimp + task_weight(cur, env->src_nid, dist) -
			      task_weight(cur, env->dst_nid, dist);
1586 1587 1588 1589 1590 1591
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1592
		} else {
1593 1594 1595 1596 1597 1598
			/*
			 * Compare the group weights. If a task is all by
			 * itself (not part of a group), use the task weight
			 * instead.
			 */
			if (cur->numa_group)
1599 1600
				imp += group_weight(cur, env->src_nid, dist) -
				       group_weight(cur, env->dst_nid, dist);
1601
			else
1602 1603
				imp += task_weight(cur, env->src_nid, dist) -
				       task_weight(cur, env->dst_nid, dist);
1604
		}
1605 1606
	}

1607
	if (imp <= env->best_imp && moveimp <= env->best_imp)
1608 1609 1610 1611
		goto unlock;

	if (!cur) {
		/* Is there capacity at our destination? */
1612
		if (env->src_stats.nr_running <= env->src_stats.task_capacity &&
1613
		    !env->dst_stats.has_free_capacity)
1614 1615 1616 1617 1618
			goto unlock;

		goto balance;
	}

1619
	/* Balance doesn't matter much if we're running a task per CPU: */
1620 1621
	if (imp > env->best_imp && src_rq->nr_running == 1 &&
			dst_rq->nr_running == 1)
1622 1623 1624 1625 1626 1627
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
1628 1629 1630
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1631

1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648
	if (moveimp > imp && moveimp > env->best_imp) {
		/*
		 * If the improvement from just moving env->p direction is
		 * better than swapping tasks around, check if a move is
		 * possible. Store a slightly smaller score than moveimp,
		 * so an actually idle CPU will win.
		 */
		if (!load_too_imbalanced(src_load, dst_load, env)) {
			imp = moveimp - 1;
			cur = NULL;
			goto assign;
		}
	}

	if (imp <= env->best_imp)
		goto unlock;

1649
	if (cur) {
1650 1651 1652
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
1653 1654
	}

1655
	if (load_too_imbalanced(src_load, dst_load, env))
1656 1657
		goto unlock;

1658 1659 1660 1661
	/*
	 * One idle CPU per node is evaluated for a task numa move.
	 * Call select_idle_sibling to maybe find a better one.
	 */
1662 1663
	if (!cur) {
		/*
1664
		 * select_idle_siblings() uses an per-CPU cpumask that
1665 1666 1667
		 * can be used from IRQ context.
		 */
		local_irq_disable();
1668 1669
		env->dst_cpu = select_idle_sibling(env->p, env->src_cpu,
						   env->dst_cpu);
1670 1671
		local_irq_enable();
	}
1672

1673 1674 1675 1676 1677 1678
assign:
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1679 1680
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1681 1682 1683 1684 1685
{
	int cpu;

	for_each_cpu(cpu, cpumask_of_node(env->dst_nid)) {
		/* Skip this CPU if the source task cannot migrate */
1686
		if (!cpumask_test_cpu(cpu, &env->p->cpus_allowed))
1687 1688 1689
			continue;

		env->dst_cpu = cpu;
1690
		task_numa_compare(env, taskimp, groupimp);
1691 1692 1693
	}
}

1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710
/* Only move tasks to a NUMA node less busy than the current node. */
static bool numa_has_capacity(struct task_numa_env *env)
{
	struct numa_stats *src = &env->src_stats;
	struct numa_stats *dst = &env->dst_stats;

	if (src->has_free_capacity && !dst->has_free_capacity)
		return false;

	/*
	 * Only consider a task move if the source has a higher load
	 * than the destination, corrected for CPU capacity on each node.
	 *
	 *      src->load                dst->load
	 * --------------------- vs ---------------------
	 * src->compute_capacity    dst->compute_capacity
	 */
1711 1712 1713
	if (src->load * dst->compute_capacity * env->imbalance_pct >

	    dst->load * src->compute_capacity * 100)
1714 1715 1716 1717 1718
		return true;

	return false;
}

1719 1720 1721 1722
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1723

1724
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1725
		.src_nid = task_node(p),
1726 1727 1728 1729 1730

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
1731
		.best_cpu = -1,
1732 1733
	};
	struct sched_domain *sd;
1734
	unsigned long taskweight, groupweight;
1735
	int nid, ret, dist;
1736
	long taskimp, groupimp;
1737

1738
	/*
1739 1740 1741 1742 1743 1744
	 * Pick the lowest SD_NUMA domain, as that would have the smallest
	 * imbalance and would be the first to start moving tasks about.
	 *
	 * And we want to avoid any moving of tasks about, as that would create
	 * random movement of tasks -- counter the numa conditions we're trying
	 * to satisfy here.
1745 1746
	 */
	rcu_read_lock();
1747
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1748 1749
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1750 1751
	rcu_read_unlock();

1752 1753 1754 1755 1756 1757 1758
	/*
	 * Cpusets can break the scheduler domain tree into smaller
	 * balance domains, some of which do not cross NUMA boundaries.
	 * Tasks that are "trapped" in such domains cannot be migrated
	 * elsewhere, so there is no point in (re)trying.
	 */
	if (unlikely(!sd)) {
1759
		p->numa_preferred_nid = task_node(p);
1760 1761 1762
		return -EINVAL;
	}

1763
	env.dst_nid = p->numa_preferred_nid;
1764 1765 1766 1767 1768 1769
	dist = env.dist = node_distance(env.src_nid, env.dst_nid);
	taskweight = task_weight(p, env.src_nid, dist);
	groupweight = group_weight(p, env.src_nid, dist);
	update_numa_stats(&env.src_stats, env.src_nid);
	taskimp = task_weight(p, env.dst_nid, dist) - taskweight;
	groupimp = group_weight(p, env.dst_nid, dist) - groupweight;
1770
	update_numa_stats(&env.dst_stats, env.dst_nid);
1771

1772
	/* Try to find a spot on the preferred nid. */
1773 1774
	if (numa_has_capacity(&env))
		task_numa_find_cpu(&env, taskimp, groupimp);
1775

1776 1777 1778 1779 1780 1781 1782
	/*
	 * Look at other nodes in these cases:
	 * - there is no space available on the preferred_nid
	 * - the task is part of a numa_group that is interleaved across
	 *   multiple NUMA nodes; in order to better consolidate the group,
	 *   we need to check other locations.
	 */
1783
	if (env.best_cpu == -1 || (p->numa_group && p->numa_group->active_nodes > 1)) {
1784 1785 1786
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1787

1788
			dist = node_distance(env.src_nid, env.dst_nid);
1789 1790 1791 1792 1793
			if (sched_numa_topology_type == NUMA_BACKPLANE &&
						dist != env.dist) {
				taskweight = task_weight(p, env.src_nid, dist);
				groupweight = group_weight(p, env.src_nid, dist);
			}
1794

1795
			/* Only consider nodes where both task and groups benefit */
1796 1797
			taskimp = task_weight(p, nid, dist) - taskweight;
			groupimp = group_weight(p, nid, dist) - groupweight;
1798
			if (taskimp < 0 && groupimp < 0)
1799 1800
				continue;

1801
			env.dist = dist;
1802 1803
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1804 1805
			if (numa_has_capacity(&env))
				task_numa_find_cpu(&env, taskimp, groupimp);
1806 1807 1808
		}
	}

1809 1810 1811 1812 1813 1814 1815 1816
	/*
	 * If the task is part of a workload that spans multiple NUMA nodes,
	 * and is migrating into one of the workload's active nodes, remember
	 * this node as the task's preferred numa node, so the workload can
	 * settle down.
	 * A task that migrated to a second choice node will be better off
	 * trying for a better one later. Do not set the preferred node here.
	 */
1817
	if (p->numa_group) {
1818 1819
		struct numa_group *ng = p->numa_group;

1820 1821 1822 1823 1824
		if (env.best_cpu == -1)
			nid = env.src_nid;
		else
			nid = env.dst_nid;

1825
		if (ng->active_nodes > 1 && numa_is_active_node(env.dst_nid, ng))
1826 1827 1828 1829 1830 1831
			sched_setnuma(p, env.dst_nid);
	}

	/* No better CPU than the current one was found. */
	if (env.best_cpu == -1)
		return -EAGAIN;
1832

1833 1834 1835 1836
	/*
	 * Reset the scan period if the task is being rescheduled on an
	 * alternative node to recheck if the tasks is now properly placed.
	 */
1837
	p->numa_scan_period = task_scan_start(p);
1838

1839
	if (env.best_task == NULL) {
1840 1841 1842
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1843 1844 1845 1846
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
1847 1848
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1849 1850
	put_task_struct(env.best_task);
	return ret;
1851 1852
}

1853 1854 1855
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1856
	unsigned long interval = HZ;
1857
	unsigned long numa_migrate_retry;
1858

1859
	/* This task has no NUMA fault statistics yet */
1860
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults))
1861 1862
		return;

1863
	/* Periodically retry migrating the task to the preferred node */
1864
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876
	numa_migrate_retry = jiffies + interval;

	/*
	 * Check that the new retry threshold is after the current one. If
	 * the retry is in the future, it implies that wake_affine has
	 * temporarily asked NUMA balancing to backoff from placement.
	 */
	if (numa_migrate_retry > p->numa_migrate_retry)
		return;

	/* Safe to try placing the task on the preferred node */
	p->numa_migrate_retry = numa_migrate_retry;
1877 1878

	/* Success if task is already running on preferred CPU */
1879
	if (task_node(p) == p->numa_preferred_nid)
1880 1881 1882
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1883
	task_numa_migrate(p);
1884 1885
}

1886
/*
1887
 * Find out how many nodes on the workload is actively running on. Do this by
1888 1889 1890 1891
 * tracking the nodes from which NUMA hinting faults are triggered. This can
 * be different from the set of nodes where the workload's memory is currently
 * located.
 */
1892
static void numa_group_count_active_nodes(struct numa_group *numa_group)
1893 1894
{
	unsigned long faults, max_faults = 0;
1895
	int nid, active_nodes = 0;
1896 1897 1898 1899 1900 1901 1902 1903 1904

	for_each_online_node(nid) {
		faults = group_faults_cpu(numa_group, nid);
		if (faults > max_faults)
			max_faults = faults;
	}

	for_each_online_node(nid) {
		faults = group_faults_cpu(numa_group, nid);
1905 1906
		if (faults * ACTIVE_NODE_FRACTION > max_faults)
			active_nodes++;
1907
	}
1908 1909 1910

	numa_group->max_faults_cpu = max_faults;
	numa_group->active_nodes = active_nodes;
1911 1912
}

1913 1914 1915
/*
 * When adapting the scan rate, the period is divided into NUMA_PERIOD_SLOTS
 * increments. The more local the fault statistics are, the higher the scan
1916 1917 1918
 * period will be for the next scan window. If local/(local+remote) ratio is
 * below NUMA_PERIOD_THRESHOLD (where range of ratio is 1..NUMA_PERIOD_SLOTS)
 * the scan period will decrease. Aim for 70% local accesses.
1919 1920
 */
#define NUMA_PERIOD_SLOTS 10
1921
#define NUMA_PERIOD_THRESHOLD 7
1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932

/*
 * Increase the scan period (slow down scanning) if the majority of
 * our memory is already on our local node, or if the majority of
 * the page accesses are shared with other processes.
 * Otherwise, decrease the scan period.
 */
static void update_task_scan_period(struct task_struct *p,
			unsigned long shared, unsigned long private)
{
	unsigned int period_slot;
1933
	int lr_ratio, ps_ratio;
1934 1935 1936 1937 1938 1939 1940 1941
	int diff;

	unsigned long remote = p->numa_faults_locality[0];
	unsigned long local = p->numa_faults_locality[1];

	/*
	 * If there were no record hinting faults then either the task is
	 * completely idle or all activity is areas that are not of interest
1942 1943 1944
	 * to automatic numa balancing. Related to that, if there were failed
	 * migration then it implies we are migrating too quickly or the local
	 * node is overloaded. In either case, scan slower
1945
	 */
1946
	if (local + shared == 0 || p->numa_faults_locality[2]) {
1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962
		p->numa_scan_period = min(p->numa_scan_period_max,
			p->numa_scan_period << 1);

		p->mm->numa_next_scan = jiffies +
			msecs_to_jiffies(p->numa_scan_period);

		return;
	}

	/*
	 * Prepare to scale scan period relative to the current period.
	 *	 == NUMA_PERIOD_THRESHOLD scan period stays the same
	 *       <  NUMA_PERIOD_THRESHOLD scan period decreases (scan faster)
	 *	 >= NUMA_PERIOD_THRESHOLD scan period increases (scan slower)
	 */
	period_slot = DIV_ROUND_UP(p->numa_scan_period, NUMA_PERIOD_SLOTS);
1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981
	lr_ratio = (local * NUMA_PERIOD_SLOTS) / (local + remote);
	ps_ratio = (private * NUMA_PERIOD_SLOTS) / (private + shared);

	if (ps_ratio >= NUMA_PERIOD_THRESHOLD) {
		/*
		 * Most memory accesses are local. There is no need to
		 * do fast NUMA scanning, since memory is already local.
		 */
		int slot = ps_ratio - NUMA_PERIOD_THRESHOLD;
		if (!slot)
			slot = 1;
		diff = slot * period_slot;
	} else if (lr_ratio >= NUMA_PERIOD_THRESHOLD) {
		/*
		 * Most memory accesses are shared with other tasks.
		 * There is no point in continuing fast NUMA scanning,
		 * since other tasks may just move the memory elsewhere.
		 */
		int slot = lr_ratio - NUMA_PERIOD_THRESHOLD;
1982 1983 1984 1985 1986
		if (!slot)
			slot = 1;
		diff = slot * period_slot;
	} else {
		/*
1987 1988 1989
		 * Private memory faults exceed (SLOTS-THRESHOLD)/SLOTS,
		 * yet they are not on the local NUMA node. Speed up
		 * NUMA scanning to get the memory moved over.
1990
		 */
1991 1992
		int ratio = max(lr_ratio, ps_ratio);
		diff = -(NUMA_PERIOD_THRESHOLD - ratio) * period_slot;
1993 1994 1995 1996 1997 1998 1999
	}

	p->numa_scan_period = clamp(p->numa_scan_period + diff,
			task_scan_min(p), task_scan_max(p));
	memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
}

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
/*
 * Get the fraction of time the task has been running since the last
 * NUMA placement cycle. The scheduler keeps similar statistics, but
 * decays those on a 32ms period, which is orders of magnitude off
 * from the dozens-of-seconds NUMA balancing period. Use the scheduler
 * stats only if the task is so new there are no NUMA statistics yet.
 */
static u64 numa_get_avg_runtime(struct task_struct *p, u64 *period)
{
	u64 runtime, delta, now;
	/* Use the start of this time slice to avoid calculations. */
	now = p->se.exec_start;
	runtime = p->se.sum_exec_runtime;

	if (p->last_task_numa_placement) {
		delta = runtime - p->last_sum_exec_runtime;
		*period = now - p->last_task_numa_placement;
	} else {
2018
		delta = p->se.avg.load_sum;
2019
		*period = LOAD_AVG_MAX;
2020 2021 2022 2023 2024 2025 2026 2027
	}

	p->last_sum_exec_runtime = runtime;
	p->last_task_numa_placement = now;

	return delta;
}

2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074
/*
 * Determine the preferred nid for a task in a numa_group. This needs to
 * be done in a way that produces consistent results with group_weight,
 * otherwise workloads might not converge.
 */
static int preferred_group_nid(struct task_struct *p, int nid)
{
	nodemask_t nodes;
	int dist;

	/* Direct connections between all NUMA nodes. */
	if (sched_numa_topology_type == NUMA_DIRECT)
		return nid;

	/*
	 * On a system with glueless mesh NUMA topology, group_weight
	 * scores nodes according to the number of NUMA hinting faults on
	 * both the node itself, and on nearby nodes.
	 */
	if (sched_numa_topology_type == NUMA_GLUELESS_MESH) {
		unsigned long score, max_score = 0;
		int node, max_node = nid;

		dist = sched_max_numa_distance;

		for_each_online_node(node) {
			score = group_weight(p, node, dist);
			if (score > max_score) {
				max_score = score;
				max_node = node;
			}
		}
		return max_node;
	}

	/*
	 * Finding the preferred nid in a system with NUMA backplane
	 * interconnect topology is more involved. The goal is to locate
	 * tasks from numa_groups near each other in the system, and
	 * untangle workloads from different sides of the system. This requires
	 * searching down the hierarchy of node groups, recursively searching
	 * inside the highest scoring group of nodes. The nodemask tricks
	 * keep the complexity of the search down.
	 */
	nodes = node_online_map;
	for (dist = sched_max_numa_distance; dist > LOCAL_DISTANCE; dist--) {
		unsigned long max_faults = 0;
2075
		nodemask_t max_group = NODE_MASK_NONE;
2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108
		int a, b;

		/* Are there nodes at this distance from each other? */
		if (!find_numa_distance(dist))
			continue;

		for_each_node_mask(a, nodes) {
			unsigned long faults = 0;
			nodemask_t this_group;
			nodes_clear(this_group);

			/* Sum group's NUMA faults; includes a==b case. */
			for_each_node_mask(b, nodes) {
				if (node_distance(a, b) < dist) {
					faults += group_faults(p, b);
					node_set(b, this_group);
					node_clear(b, nodes);
				}
			}

			/* Remember the top group. */
			if (faults > max_faults) {
				max_faults = faults;
				max_group = this_group;
				/*
				 * subtle: at the smallest distance there is
				 * just one node left in each "group", the
				 * winner is the preferred nid.
				 */
				nid = a;
			}
		}
		/* Next round, evaluate the nodes within max_group. */
2109 2110
		if (!max_faults)
			break;
2111 2112 2113 2114 2115
		nodes = max_group;
	}
	return nid;
}

2116 2117
static void task_numa_placement(struct task_struct *p)
{
2118 2119
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
2120
	unsigned long fault_types[2] = { 0, 0 };
2121 2122
	unsigned long total_faults;
	u64 runtime, period;
2123
	spinlock_t *group_lock = NULL;
2124

2125 2126 2127 2128 2129
	/*
	 * The p->mm->numa_scan_seq field gets updated without
	 * exclusive access. Use READ_ONCE() here to ensure
	 * that the field is read in a single access:
	 */
2130
	seq = READ_ONCE(p->mm->numa_scan_seq);
2131 2132 2133
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
2134
	p->numa_scan_period_max = task_scan_max(p);
2135

2136 2137 2138 2139
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

2140 2141 2142
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
2143
		spin_lock_irq(group_lock);
2144 2145
	}

2146 2147
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
2148 2149
		/* Keep track of the offsets in numa_faults array */
		int mem_idx, membuf_idx, cpu_idx, cpubuf_idx;
2150
		unsigned long faults = 0, group_faults = 0;
2151
		int priv;
2152

2153
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
2154
			long diff, f_diff, f_weight;
2155

2156 2157 2158 2159
			mem_idx = task_faults_idx(NUMA_MEM, nid, priv);
			membuf_idx = task_faults_idx(NUMA_MEMBUF, nid, priv);
			cpu_idx = task_faults_idx(NUMA_CPU, nid, priv);
			cpubuf_idx = task_faults_idx(NUMA_CPUBUF, nid, priv);
2160

2161
			/* Decay existing window, copy faults since last scan */
2162 2163 2164
			diff = p->numa_faults[membuf_idx] - p->numa_faults[mem_idx] / 2;
			fault_types[priv] += p->numa_faults[membuf_idx];
			p->numa_faults[membuf_idx] = 0;
2165

2166 2167 2168 2169 2170 2171 2172 2173
			/*
			 * Normalize the faults_from, so all tasks in a group
			 * count according to CPU use, instead of by the raw
			 * number of faults. Tasks with little runtime have
			 * little over-all impact on throughput, and thus their
			 * faults are less important.
			 */
			f_weight = div64_u64(runtime << 16, period + 1);
2174
			f_weight = (f_weight * p->numa_faults[cpubuf_idx]) /
2175
				   (total_faults + 1);
2176 2177
			f_diff = f_weight - p->numa_faults[cpu_idx] / 2;
			p->numa_faults[cpubuf_idx] = 0;
2178

2179 2180 2181
			p->numa_faults[mem_idx] += diff;
			p->numa_faults[cpu_idx] += f_diff;
			faults += p->numa_faults[mem_idx];
2182
			p->total_numa_faults += diff;
2183
			if (p->numa_group) {
2184 2185 2186 2187 2188 2189 2190 2191 2192
				/*
				 * safe because we can only change our own group
				 *
				 * mem_idx represents the offset for a given
				 * nid and priv in a specific region because it
				 * is at the beginning of the numa_faults array.
				 */
				p->numa_group->faults[mem_idx] += diff;
				p->numa_group->faults_cpu[mem_idx] += f_diff;
2193
				p->numa_group->total_faults += diff;
2194
				group_faults += p->numa_group->faults[mem_idx];
2195
			}
2196 2197
		}

2198 2199 2200 2201
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
2202 2203 2204 2205 2206 2207 2208

		if (group_faults > max_group_faults) {
			max_group_faults = group_faults;
			max_group_nid = nid;
		}
	}

2209 2210
	update_task_scan_period(p, fault_types[0], fault_types[1]);

2211
	if (p->numa_group) {
2212
		numa_group_count_active_nodes(p->numa_group);
2213
		spin_unlock_irq(group_lock);
2214
		max_nid = preferred_group_nid(p, max_group_nid);
2215 2216
	}

2217 2218 2219 2220 2221 2222 2223
	if (max_faults) {
		/* Set the new preferred node */
		if (max_nid != p->numa_preferred_nid)
			sched_setnuma(p, max_nid);

		if (task_node(p) != p->numa_preferred_nid)
			numa_migrate_preferred(p);
2224
	}
2225 2226
}

2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237
static inline int get_numa_group(struct numa_group *grp)
{
	return atomic_inc_not_zero(&grp->refcount);
}

static inline void put_numa_group(struct numa_group *grp)
{
	if (atomic_dec_and_test(&grp->refcount))
		kfree_rcu(grp, rcu);
}

2238 2239
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
2240 2241 2242 2243 2244 2245 2246 2247 2248
{
	struct numa_group *grp, *my_grp;
	struct task_struct *tsk;
	bool join = false;
	int cpu = cpupid_to_cpu(cpupid);
	int i;

	if (unlikely(!p->numa_group)) {
		unsigned int size = sizeof(struct numa_group) +
2249
				    4*nr_node_ids*sizeof(unsigned long);
2250 2251 2252 2253 2254 2255

		grp = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
		if (!grp)
			return;

		atomic_set(&grp->refcount, 1);
2256 2257
		grp->active_nodes = 1;
		grp->max_faults_cpu = 0;
2258
		spin_lock_init(&grp->lock);
2259
		grp->gid = p->pid;
2260
		/* Second half of the array tracks nids where faults happen */
2261 2262
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
2263

2264
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2265
			grp->faults[i] = p->numa_faults[i];
2266

2267
		grp->total_faults = p->total_numa_faults;
2268

2269 2270 2271 2272 2273
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
2274
	tsk = READ_ONCE(cpu_rq(cpu)->curr);
2275 2276

	if (!cpupid_match_pid(tsk, cpupid))
2277
		goto no_join;
2278 2279 2280

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
2281
		goto no_join;
2282 2283 2284

	my_grp = p->numa_group;
	if (grp == my_grp)
2285
		goto no_join;
2286 2287 2288 2289 2290 2291

	/*
	 * Only join the other group if its bigger; if we're the bigger group,
	 * the other task will join us.
	 */
	if (my_grp->nr_tasks > grp->nr_tasks)
2292
		goto no_join;
2293 2294 2295 2296 2297

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
2298
		goto no_join;
2299

2300 2301 2302 2303 2304 2305 2306
	/* Always join threads in the same process. */
	if (tsk->mm == current->mm)
		join = true;

	/* Simple filter to avoid false positives due to PID collisions */
	if (flags & TNF_SHARED)
		join = true;
2307

2308 2309 2310
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

2311
	if (join && !get_numa_group(grp))
2312
		goto no_join;
2313 2314 2315 2316 2317 2318

	rcu_read_unlock();

	if (!join)
		return;

2319 2320
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
2321

2322
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
2323 2324
		my_grp->faults[i] -= p->numa_faults[i];
		grp->faults[i] += p->numa_faults[i];
2325
	}
2326 2327
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
2328 2329 2330 2331 2332

	my_grp->nr_tasks--;
	grp->nr_tasks++;

	spin_unlock(&my_grp->lock);
2333
	spin_unlock_irq(&grp->lock);
2334 2335 2336 2337

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
2338 2339 2340 2341 2342
	return;

no_join:
	rcu_read_unlock();
	return;
2343 2344 2345 2346 2347
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
2348
	void *numa_faults = p->numa_faults;
2349 2350
	unsigned long flags;
	int i;
2351 2352

	if (grp) {
2353
		spin_lock_irqsave(&grp->lock, flags);
2354
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2355
			grp->faults[i] -= p->numa_faults[i];
2356
		grp->total_faults -= p->total_numa_faults;
2357

2358
		grp->nr_tasks--;
2359
		spin_unlock_irqrestore(&grp->lock, flags);
2360
		RCU_INIT_POINTER(p->numa_group, NULL);
2361 2362 2363
		put_numa_group(grp);
	}

2364
	p->numa_faults = NULL;
2365
	kfree(numa_faults);
2366 2367
}

2368 2369 2370
/*
 * Got a PROT_NONE fault for a page on @node.
 */
2371
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
2372 2373
{
	struct task_struct *p = current;
2374
	bool migrated = flags & TNF_MIGRATED;
2375
	int cpu_node = task_node(current);
2376
	int local = !!(flags & TNF_FAULT_LOCAL);
2377
	struct numa_group *ng;
2378
	int priv;
2379

2380
	if (!static_branch_likely(&sched_numa_balancing))
2381 2382
		return;

2383 2384 2385 2386
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

2387
	/* Allocate buffer to track faults on a per-node basis */
2388 2389
	if (unlikely(!p->numa_faults)) {
		int size = sizeof(*p->numa_faults) *
2390
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
2391

2392 2393
		p->numa_faults = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
		if (!p->numa_faults)
2394
			return;
2395

2396
		p->total_numa_faults = 0;
2397
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
2398
	}
2399

2400 2401 2402 2403 2404 2405 2406 2407
	/*
	 * First accesses are treated as private, otherwise consider accesses
	 * to be private if the accessing pid has not changed
	 */
	if (unlikely(last_cpupid == (-1 & LAST_CPUPID_MASK))) {
		priv = 1;
	} else {
		priv = cpupid_match_pid(p, last_cpupid);
2408
		if (!priv && !(flags & TNF_NO_GROUP))
2409
			task_numa_group(p, last_cpupid, flags, &priv);
2410 2411
	}

2412 2413 2414 2415 2416 2417
	/*
	 * If a workload spans multiple NUMA nodes, a shared fault that
	 * occurs wholly within the set of nodes that the workload is
	 * actively using should be counted as local. This allows the
	 * scan rate to slow down when a workload has settled down.
	 */
2418 2419 2420 2421
	ng = p->numa_group;
	if (!priv && !local && ng && ng->active_nodes > 1 &&
				numa_is_active_node(cpu_node, ng) &&
				numa_is_active_node(mem_node, ng))
2422 2423
		local = 1;

2424
	task_numa_placement(p);
2425

2426 2427 2428 2429 2430
	/*
	 * Retry task to preferred node migration periodically, in case it
	 * case it previously failed, or the scheduler moved us.
	 */
	if (time_after(jiffies, p->numa_migrate_retry))
2431 2432
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
2433 2434
	if (migrated)
		p->numa_pages_migrated += pages;
2435 2436
	if (flags & TNF_MIGRATE_FAIL)
		p->numa_faults_locality[2] += pages;
I
Ingo Molnar 已提交
2437

2438 2439
	p->numa_faults[task_faults_idx(NUMA_MEMBUF, mem_node, priv)] += pages;
	p->numa_faults[task_faults_idx(NUMA_CPUBUF, cpu_node, priv)] += pages;
2440
	p->numa_faults_locality[local] += pages;
2441 2442
}

2443 2444
static void reset_ptenuma_scan(struct task_struct *p)
{
2445 2446 2447 2448 2449 2450 2451 2452
	/*
	 * We only did a read acquisition of the mmap sem, so
	 * p->mm->numa_scan_seq is written to without exclusive access
	 * and the update is not guaranteed to be atomic. That's not
	 * much of an issue though, since this is just used for
	 * statistical sampling. Use READ_ONCE/WRITE_ONCE, which are not
	 * expensive, to avoid any form of compiler optimizations:
	 */
2453
	WRITE_ONCE(p->mm->numa_scan_seq, READ_ONCE(p->mm->numa_scan_seq) + 1);
2454 2455 2456
	p->mm->numa_scan_offset = 0;
}

2457 2458 2459 2460 2461 2462 2463 2464 2465
/*
 * The expensive part of numa migration is done from task_work context.
 * Triggered from task_tick_numa().
 */
void task_numa_work(struct callback_head *work)
{
	unsigned long migrate, next_scan, now = jiffies;
	struct task_struct *p = current;
	struct mm_struct *mm = p->mm;
2466
	u64 runtime = p->se.sum_exec_runtime;
2467
	struct vm_area_struct *vma;
2468
	unsigned long start, end;
2469
	unsigned long nr_pte_updates = 0;
2470
	long pages, virtpages;
2471

2472
	SCHED_WARN_ON(p != container_of(work, struct task_struct, numa_work));
2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485

	work->next = work; /* protect against double add */
	/*
	 * Who cares about NUMA placement when they're dying.
	 *
	 * NOTE: make sure not to dereference p->mm before this check,
	 * exit_task_work() happens _after_ exit_mm() so we could be called
	 * without p->mm even though we still had it when we enqueued this
	 * work.
	 */
	if (p->flags & PF_EXITING)
		return;

2486
	if (!mm->numa_next_scan) {
2487 2488
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2489 2490
	}

2491 2492 2493 2494 2495 2496 2497
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

2498 2499
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
2500
		p->numa_scan_period = task_scan_start(p);
2501
	}
2502

2503
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
2504 2505 2506
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

2507 2508 2509 2510 2511 2512
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

2513 2514 2515
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
2516
	virtpages = pages * 8;	   /* Scan up to this much virtual space */
2517 2518
	if (!pages)
		return;
2519

2520

2521 2522
	if (!down_read_trylock(&mm->mmap_sem))
		return;
2523
	vma = find_vma(mm, start);
2524 2525
	if (!vma) {
		reset_ptenuma_scan(p);
2526
		start = 0;
2527 2528
		vma = mm->mmap;
	}
2529
	for (; vma; vma = vma->vm_next) {
2530
		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
2531
			is_vm_hugetlb_page(vma) || (vma->vm_flags & VM_MIXEDMAP)) {
2532
			continue;
2533
		}
2534

2535 2536 2537 2538 2539 2540 2541 2542 2543 2544
		/*
		 * Shared library pages mapped by multiple processes are not
		 * migrated as it is expected they are cache replicated. Avoid
		 * hinting faults in read-only file-backed mappings or the vdso
		 * as migrating the pages will be of marginal benefit.
		 */
		if (!vma->vm_mm ||
		    (vma->vm_file && (vma->vm_flags & (VM_READ|VM_WRITE)) == (VM_READ)))
			continue;

M
Mel Gorman 已提交
2545 2546 2547 2548 2549 2550
		/*
		 * Skip inaccessible VMAs to avoid any confusion between
		 * PROT_NONE and NUMA hinting ptes
		 */
		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
			continue;
2551

2552 2553 2554 2555
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2556
			nr_pte_updates = change_prot_numa(vma, start, end);
2557 2558

			/*
2559 2560 2561 2562 2563 2564
			 * Try to scan sysctl_numa_balancing_size worth of
			 * hpages that have at least one present PTE that
			 * is not already pte-numa. If the VMA contains
			 * areas that are unused or already full of prot_numa
			 * PTEs, scan up to virtpages, to skip through those
			 * areas faster.
2565 2566 2567
			 */
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
2568
			virtpages -= (end - start) >> PAGE_SHIFT;
2569

2570
			start = end;
2571
			if (pages <= 0 || virtpages <= 0)
2572
				goto out;
2573 2574

			cond_resched();
2575
		} while (end != vma->vm_end);
2576
	}
2577

2578
out:
2579
	/*
P
Peter Zijlstra 已提交
2580 2581 2582 2583
	 * It is possible to reach the end of the VMA list but the last few
	 * VMAs are not guaranteed to the vma_migratable. If they are not, we
	 * would find the !migratable VMA on the next scan but not reset the
	 * scanner to the start so check it now.
2584 2585
	 */
	if (vma)
2586
		mm->numa_scan_offset = start;
2587 2588 2589
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600

	/*
	 * Make sure tasks use at least 32x as much time to run other code
	 * than they used here, to limit NUMA PTE scanning overhead to 3% max.
	 * Usually update_task_scan_period slows down scanning enough; on an
	 * overloaded system we need to limit overhead on a per task basis.
	 */
	if (unlikely(p->se.sum_exec_runtime != runtime)) {
		u64 diff = p->se.sum_exec_runtime - runtime;
		p->node_stamp += 32 * diff;
	}
2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625
}

/*
 * Drive the periodic memory faults..
 */
void task_tick_numa(struct rq *rq, struct task_struct *curr)
{
	struct callback_head *work = &curr->numa_work;
	u64 period, now;

	/*
	 * We don't care about NUMA placement if we don't have memory.
	 */
	if (!curr->mm || (curr->flags & PF_EXITING) || work->next != work)
		return;

	/*
	 * Using runtime rather than walltime has the dual advantage that
	 * we (mostly) drive the selection from busy threads and that the
	 * task needs to have done some actual work before we bother with
	 * NUMA placement.
	 */
	now = curr->se.sum_exec_runtime;
	period = (u64)curr->numa_scan_period * NSEC_PER_MSEC;

2626
	if (now > curr->node_stamp + period) {
2627
		if (!curr->node_stamp)
2628
			curr->numa_scan_period = task_scan_start(curr);
2629
		curr->node_stamp += period;
2630 2631 2632 2633 2634 2635 2636

		if (!time_before(jiffies, curr->mm->numa_next_scan)) {
			init_task_work(work, task_numa_work); /* TODO: move this into sched_fork() */
			task_work_add(curr, work, true);
		}
	}
}
2637

2638 2639 2640 2641
#else
static void task_tick_numa(struct rq *rq, struct task_struct *curr)
{
}
2642 2643 2644 2645 2646 2647 2648 2649

static inline void account_numa_enqueue(struct rq *rq, struct task_struct *p)
{
}

static inline void account_numa_dequeue(struct rq *rq, struct task_struct *p)
{
}
2650

2651 2652
#endif /* CONFIG_NUMA_BALANCING */

2653 2654 2655 2656
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2657
	if (!parent_entity(se))
2658
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2659
#ifdef CONFIG_SMP
2660 2661 2662 2663 2664 2665
	if (entity_is_task(se)) {
		struct rq *rq = rq_of(cfs_rq);

		account_numa_enqueue(rq, task_of(se));
		list_add(&se->group_node, &rq->cfs_tasks);
	}
2666
#endif
2667 2668 2669 2670 2671 2672 2673
	cfs_rq->nr_running++;
}

static void
account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_sub(&cfs_rq->load, se->load.weight);
2674
	if (!parent_entity(se))
2675
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2676
#ifdef CONFIG_SMP
2677 2678
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2679
		list_del_init(&se->group_node);
2680
	}
2681
#endif
2682 2683 2684
	cfs_rq->nr_running--;
}

2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723
/*
 * Signed add and clamp on underflow.
 *
 * Explicitly do a load-store to ensure the intermediate value never hits
 * memory. This allows lockless observations without ever seeing the negative
 * values.
 */
#define add_positive(_ptr, _val) do {                           \
	typeof(_ptr) ptr = (_ptr);                              \
	typeof(_val) val = (_val);                              \
	typeof(*ptr) res, var = READ_ONCE(*ptr);                \
								\
	res = var + val;                                        \
								\
	if (val < 0 && res > var)                               \
		res = 0;                                        \
								\
	WRITE_ONCE(*ptr, res);                                  \
} while (0)

/*
 * Unsigned subtract and clamp on underflow.
 *
 * Explicitly do a load-store to ensure the intermediate value never hits
 * memory. This allows lockless observations without ever seeing the negative
 * values.
 */
#define sub_positive(_ptr, _val) do {				\
	typeof(_ptr) ptr = (_ptr);				\
	typeof(*ptr) val = (_val);				\
	typeof(*ptr) res, var = READ_ONCE(*ptr);		\
	res = var - val;					\
	if (res > var)						\
		res = 0;					\
	WRITE_ONCE(*ptr, res);					\
} while (0)

#ifdef CONFIG_SMP
/*
2724
 * XXX we want to get rid of these helpers and use the full load resolution.
2725 2726 2727 2728 2729 2730
 */
static inline long se_weight(struct sched_entity *se)
{
	return scale_load_down(se->load.weight);
}

2731 2732 2733 2734 2735
static inline long se_runnable(struct sched_entity *se)
{
	return scale_load_down(se->runnable_weight);
}

2736 2737 2738
static inline void
enqueue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2739 2740 2741 2742
	cfs_rq->runnable_weight += se->runnable_weight;

	cfs_rq->avg.runnable_load_avg += se->avg.runnable_load_avg;
	cfs_rq->avg.runnable_load_sum += se_runnable(se) * se->avg.runnable_load_sum;
2743 2744 2745 2746 2747
}

static inline void
dequeue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2748 2749 2750 2751 2752
	cfs_rq->runnable_weight -= se->runnable_weight;

	sub_positive(&cfs_rq->avg.runnable_load_avg, se->avg.runnable_load_avg);
	sub_positive(&cfs_rq->avg.runnable_load_sum,
		     se_runnable(se) * se->avg.runnable_load_sum);
2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778
}

static inline void
enqueue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	cfs_rq->avg.load_avg += se->avg.load_avg;
	cfs_rq->avg.load_sum += se_weight(se) * se->avg.load_sum;
}

static inline void
dequeue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	sub_positive(&cfs_rq->avg.load_avg, se->avg.load_avg);
	sub_positive(&cfs_rq->avg.load_sum, se_weight(se) * se->avg.load_sum);
}
#else
static inline void
enqueue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) { }
static inline void
dequeue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) { }
static inline void
enqueue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) { }
static inline void
dequeue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) { }
#endif

2779
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
2780
			    unsigned long weight, unsigned long runnable)
2781 2782 2783 2784 2785 2786 2787 2788 2789 2790
{
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
		account_entity_dequeue(cfs_rq, se);
		dequeue_runnable_load_avg(cfs_rq, se);
	}
	dequeue_load_avg(cfs_rq, se);

2791
	se->runnable_weight = runnable;
2792 2793 2794
	update_load_set(&se->load, weight);

#ifdef CONFIG_SMP
2795 2796 2797 2798 2799 2800 2801
	do {
		u32 divider = LOAD_AVG_MAX - 1024 + se->avg.period_contrib;

		se->avg.load_avg = div_u64(se_weight(se) * se->avg.load_sum, divider);
		se->avg.runnable_load_avg =
			div_u64(se_runnable(se) * se->avg.runnable_load_sum, divider);
	} while (0);
2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817
#endif

	enqueue_load_avg(cfs_rq, se);
	if (se->on_rq) {
		account_entity_enqueue(cfs_rq, se);
		enqueue_runnable_load_avg(cfs_rq, se);
	}
}

void reweight_task(struct task_struct *p, int prio)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	struct load_weight *load = &se->load;
	unsigned long weight = scale_load(sched_prio_to_weight[prio]);

2818
	reweight_entity(cfs_rq, se, weight, weight);
2819 2820 2821
	load->inv_weight = sched_prio_to_wmult[prio];
}

2822
#ifdef CONFIG_FAIR_GROUP_SCHED
2823
#ifdef CONFIG_SMP
2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861
/*
 * All this does is approximate the hierarchical proportion which includes that
 * global sum we all love to hate.
 *
 * That is, the weight of a group entity, is the proportional share of the
 * group weight based on the group runqueue weights. That is:
 *
 *                     tg->weight * grq->load.weight
 *   ge->load.weight = -----------------------------               (1)
 *			  \Sum grq->load.weight
 *
 * Now, because computing that sum is prohibitively expensive to compute (been
 * there, done that) we approximate it with this average stuff. The average
 * moves slower and therefore the approximation is cheaper and more stable.
 *
 * So instead of the above, we substitute:
 *
 *   grq->load.weight -> grq->avg.load_avg                         (2)
 *
 * which yields the following:
 *
 *                     tg->weight * grq->avg.load_avg
 *   ge->load.weight = ------------------------------              (3)
 *				tg->load_avg
 *
 * Where: tg->load_avg ~= \Sum grq->avg.load_avg
 *
 * That is shares_avg, and it is right (given the approximation (2)).
 *
 * The problem with it is that because the average is slow -- it was designed
 * to be exactly that of course -- this leads to transients in boundary
 * conditions. In specific, the case where the group was idle and we start the
 * one task. It takes time for our CPU's grq->avg.load_avg to build up,
 * yielding bad latency etc..
 *
 * Now, in that special case (1) reduces to:
 *
 *                     tg->weight * grq->load.weight
2862
 *   ge->load.weight = ----------------------------- = tg->weight   (4)
2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875
 *			    grp->load.weight
 *
 * That is, the sum collapses because all other CPUs are idle; the UP scenario.
 *
 * So what we do is modify our approximation (3) to approach (4) in the (near)
 * UP case, like:
 *
 *   ge->load.weight =
 *
 *              tg->weight * grq->load.weight
 *     ---------------------------------------------------         (5)
 *     tg->load_avg - grq->avg.load_avg + grq->load.weight
 *
2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887
 * But because grq->load.weight can drop to 0, resulting in a divide by zero,
 * we need to use grq->avg.load_avg as its lower bound, which then gives:
 *
 *
 *                     tg->weight * grq->load.weight
 *   ge->load.weight = -----------------------------		   (6)
 *				tg_load_avg'
 *
 * Where:
 *
 *   tg_load_avg' = tg->load_avg - grq->avg.load_avg +
 *                  max(grq->load.weight, grq->avg.load_avg)
2888 2889 2890 2891 2892 2893 2894 2895 2896
 *
 * And that is shares_weight and is icky. In the (near) UP case it approaches
 * (4) while in the normal case it approaches (3). It consistently
 * overestimates the ge->load.weight and therefore:
 *
 *   \Sum ge->load.weight >= tg->weight
 *
 * hence icky!
 */
2897
static long calc_group_shares(struct cfs_rq *cfs_rq)
2898
{
2899 2900 2901 2902
	long tg_weight, tg_shares, load, shares;
	struct task_group *tg = cfs_rq->tg;

	tg_shares = READ_ONCE(tg->shares);
2903

2904
	load = max(scale_load_down(cfs_rq->load.weight), cfs_rq->avg.load_avg);
2905

2906
	tg_weight = atomic_long_read(&tg->load_avg);
2907

2908 2909 2910
	/* Ensure tg_weight >= load */
	tg_weight -= cfs_rq->tg_load_avg_contrib;
	tg_weight += load;
2911

2912
	shares = (tg_shares * load);
2913 2914
	if (tg_weight)
		shares /= tg_weight;
2915

2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927
	/*
	 * MIN_SHARES has to be unscaled here to support per-CPU partitioning
	 * of a group with small tg->shares value. It is a floor value which is
	 * assigned as a minimum load.weight to the sched_entity representing
	 * the group on a CPU.
	 *
	 * E.g. on 64-bit for a group with tg->shares of scale_load(15)=15*1024
	 * on an 8-core system with 8 tasks each runnable on one CPU shares has
	 * to be 15*1024*1/8=1920 instead of scale_load(MIN_SHARES)=2*1024. In
	 * case no task is runnable on a CPU MIN_SHARES=2 should be returned
	 * instead of 0.
	 */
2928
	return clamp_t(long, shares, MIN_SHARES, tg_shares);
2929
}
2930 2931

/*
2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956
 * This calculates the effective runnable weight for a group entity based on
 * the group entity weight calculated above.
 *
 * Because of the above approximation (2), our group entity weight is
 * an load_avg based ratio (3). This means that it includes blocked load and
 * does not represent the runnable weight.
 *
 * Approximate the group entity's runnable weight per ratio from the group
 * runqueue:
 *
 *					     grq->avg.runnable_load_avg
 *   ge->runnable_weight = ge->load.weight * -------------------------- (7)
 *						 grq->avg.load_avg
 *
 * However, analogous to above, since the avg numbers are slow, this leads to
 * transients in the from-idle case. Instead we use:
 *
 *   ge->runnable_weight = ge->load.weight *
 *
 *		max(grq->avg.runnable_load_avg, grq->runnable_weight)
 *		-----------------------------------------------------	(8)
 *		      max(grq->avg.load_avg, grq->load.weight)
 *
 * Where these max() serve both to use the 'instant' values to fix the slow
 * from-idle and avoid the /0 on to-idle, similar to (6).
2957 2958 2959
 */
static long calc_group_runnable(struct cfs_rq *cfs_rq, long shares)
{
2960 2961 2962 2963 2964 2965 2966
	long runnable, load_avg;

	load_avg = max(cfs_rq->avg.load_avg,
		       scale_load_down(cfs_rq->load.weight));

	runnable = max(cfs_rq->avg.runnable_load_avg,
		       scale_load_down(cfs_rq->runnable_weight));
2967 2968 2969 2970

	runnable *= shares;
	if (load_avg)
		runnable /= load_avg;
2971

2972 2973
	return clamp_t(long, runnable, MIN_SHARES, shares);
}
2974
#endif /* CONFIG_SMP */
2975

2976 2977
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2978 2979 2980 2981 2982
/*
 * Recomputes the group entity based on the current state of its group
 * runqueue.
 */
static void update_cfs_group(struct sched_entity *se)
P
Peter Zijlstra 已提交
2983
{
2984 2985
	struct cfs_rq *gcfs_rq = group_cfs_rq(se);
	long shares, runnable;
P
Peter Zijlstra 已提交
2986

2987
	if (!gcfs_rq)
2988 2989
		return;

2990
	if (throttled_hierarchy(gcfs_rq))
P
Peter Zijlstra 已提交
2991
		return;
2992

2993
#ifndef CONFIG_SMP
2994
	runnable = shares = READ_ONCE(gcfs_rq->tg->shares);
2995 2996

	if (likely(se->load.weight == shares))
2997
		return;
2998
#else
2999 3000
	shares   = calc_group_shares(gcfs_rq);
	runnable = calc_group_runnable(gcfs_rq, shares);
3001
#endif
P
Peter Zijlstra 已提交
3002

3003
	reweight_entity(cfs_rq_of(se), se, shares, runnable);
P
Peter Zijlstra 已提交
3004
}
3005

P
Peter Zijlstra 已提交
3006
#else /* CONFIG_FAIR_GROUP_SCHED */
3007
static inline void update_cfs_group(struct sched_entity *se)
P
Peter Zijlstra 已提交
3008 3009 3010 3011
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

3012
static inline void cfs_rq_util_change(struct cfs_rq *cfs_rq, int flags)
3013
{
3014 3015
	struct rq *rq = rq_of(cfs_rq);

3016
	if (&rq->cfs == cfs_rq || (flags & SCHED_CPUFREQ_MIGRATION)) {
3017 3018 3019
		/*
		 * There are a few boundary cases this might miss but it should
		 * get called often enough that that should (hopefully) not be
3020
		 * a real problem.
3021 3022 3023 3024 3025 3026 3027 3028 3029 3030
		 *
		 * It will not get called when we go idle, because the idle
		 * thread is a different class (!fair), nor will the utilization
		 * number include things like RT tasks.
		 *
		 * As is, the util number is not freq-invariant (we'd have to
		 * implement arch_scale_freq_capacity() for that).
		 *
		 * See cpu_util().
		 */
3031
		cpufreq_update_util(rq, flags);
3032 3033 3034
	}
}

3035
#ifdef CONFIG_SMP
3036 3037 3038 3039
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
3040
static u64 decay_load(u64 val, u64 n)
3041
{
3042 3043
	unsigned int local_n;

3044
	if (unlikely(n > LOAD_AVG_PERIOD * 63))
3045 3046 3047 3048 3049 3050 3051
		return 0;

	/* after bounds checking we can collapse to 32-bit */
	local_n = n;

	/*
	 * As y^PERIOD = 1/2, we can combine
3052 3053
	 *    y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
	 * With a look-up table which covers y^n (n<PERIOD)
3054 3055 3056 3057 3058 3059
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
3060 3061
	}

3062 3063
	val = mul_u64_u32_shr(val, runnable_avg_yN_inv[local_n], 32);
	return val;
3064 3065
}

3066
static u32 __accumulate_pelt_segments(u64 periods, u32 d1, u32 d3)
3067
{
3068
	u32 c1, c2, c3 = d3; /* y^0 == 1 */
3069

3070
	/*
P
Peter Zijlstra 已提交
3071
	 * c1 = d1 y^p
3072
	 */
3073
	c1 = decay_load((u64)d1, periods);
3074 3075

	/*
P
Peter Zijlstra 已提交
3076
	 *            p-1
3077 3078
	 * c2 = 1024 \Sum y^n
	 *            n=1
3079
	 *
3080 3081
	 *              inf        inf
	 *    = 1024 ( \Sum y^n - \Sum y^n - y^0 )
P
Peter Zijlstra 已提交
3082
	 *              n=0        n=p
3083
	 */
3084
	c2 = LOAD_AVG_MAX - decay_load(LOAD_AVG_MAX, periods) - 1024;
3085 3086

	return c1 + c2 + c3;
3087 3088
}

3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099
/*
 * Accumulate the three separate parts of the sum; d1 the remainder
 * of the last (incomplete) period, d2 the span of full periods and d3
 * the remainder of the (incomplete) current period.
 *
 *           d1          d2           d3
 *           ^           ^            ^
 *           |           |            |
 *         |<->|<----------------->|<--->|
 * ... |---x---|------| ... |------|-----x (now)
 *
P
Peter Zijlstra 已提交
3100 3101 3102
 *                           p-1
 * u' = (u + d1) y^p + 1024 \Sum y^n + d3 y^0
 *                           n=1
3103
 *
P
Peter Zijlstra 已提交
3104
 *    = u y^p +					(Step 1)
3105
 *
P
Peter Zijlstra 已提交
3106 3107 3108
 *                     p-1
 *      d1 y^p + 1024 \Sum y^n + d3 y^0		(Step 2)
 *                     n=1
3109 3110 3111
 */
static __always_inline u32
accumulate_sum(u64 delta, int cpu, struct sched_avg *sa,
3112
	       unsigned long load, unsigned long runnable, int running)
3113 3114
{
	unsigned long scale_freq, scale_cpu;
3115
	u32 contrib = (u32)delta; /* p == 0 -> delta < 1024 */
3116 3117
	u64 periods;

3118
	scale_freq = arch_scale_freq_capacity(cpu);
3119 3120 3121 3122 3123 3124 3125 3126 3127 3128
	scale_cpu = arch_scale_cpu_capacity(NULL, cpu);

	delta += sa->period_contrib;
	periods = delta / 1024; /* A period is 1024us (~1ms) */

	/*
	 * Step 1: decay old *_sum if we crossed period boundaries.
	 */
	if (periods) {
		sa->load_sum = decay_load(sa->load_sum, periods);
3129 3130
		sa->runnable_load_sum =
			decay_load(sa->runnable_load_sum, periods);
3131 3132
		sa->util_sum = decay_load((u64)(sa->util_sum), periods);

3133 3134 3135 3136 3137 3138 3139
		/*
		 * Step 2
		 */
		delta %= 1024;
		contrib = __accumulate_pelt_segments(periods,
				1024 - sa->period_contrib, delta);
	}
3140 3141 3142
	sa->period_contrib = delta;

	contrib = cap_scale(contrib, scale_freq);
3143 3144 3145 3146
	if (load)
		sa->load_sum += load * contrib;
	if (runnable)
		sa->runnable_load_sum += runnable * contrib;
3147 3148 3149 3150 3151 3152
	if (running)
		sa->util_sum += contrib * scale_cpu;

	return periods;
}

3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180
/*
 * We can represent the historical contribution to runnable average as the
 * coefficients of a geometric series.  To do this we sub-divide our runnable
 * history into segments of approximately 1ms (1024us); label the segment that
 * occurred N-ms ago p_N, with p_0 corresponding to the current period, e.g.
 *
 * [<- 1024us ->|<- 1024us ->|<- 1024us ->| ...
 *      p0            p1           p2
 *     (now)       (~1ms ago)  (~2ms ago)
 *
 * Let u_i denote the fraction of p_i that the entity was runnable.
 *
 * We then designate the fractions u_i as our co-efficients, yielding the
 * following representation of historical load:
 *   u_0 + u_1*y + u_2*y^2 + u_3*y^3 + ...
 *
 * We choose y based on the with of a reasonably scheduling period, fixing:
 *   y^32 = 0.5
 *
 * This means that the contribution to load ~32ms ago (u_32) will be weighted
 * approximately half as much as the contribution to load within the last ms
 * (u_0).
 *
 * When a period "rolls over" and we have new u_0`, multiplying the previous
 * sum again by y is sufficient to update:
 *   load_avg = u_0` + y*(u_0 + u_1*y + u_2*y^2 + ... )
 *            = u_0 + u_1*y + u_2*y^2 + ... [re-labeling u_i --> u_{i+1}]
 */
3181
static __always_inline int
3182
___update_load_sum(u64 now, int cpu, struct sched_avg *sa,
3183
		  unsigned long load, unsigned long runnable, int running)
3184
{
3185
	u64 delta;
3186

3187
	delta = now - sa->last_update_time;
3188 3189 3190 3191 3192
	/*
	 * This should only happen when time goes backwards, which it
	 * unfortunately does during sched clock init when we swap over to TSC.
	 */
	if ((s64)delta < 0) {
3193
		sa->last_update_time = now;
3194 3195 3196 3197 3198 3199 3200 3201 3202 3203
		return 0;
	}

	/*
	 * Use 1024ns as the unit of measurement since it's a reasonable
	 * approximation of 1us and fast to compute.
	 */
	delta >>= 10;
	if (!delta)
		return 0;
3204 3205

	sa->last_update_time += delta << 10;
3206

3207 3208 3209 3210 3211 3212 3213 3214 3215
	/*
	 * running is a subset of runnable (weight) so running can't be set if
	 * runnable is clear. But there are some corner cases where the current
	 * se has been already dequeued but cfs_rq->curr still points to it.
	 * This means that weight will be 0 but not running for a sched_entity
	 * but also for a cfs_rq if the latter becomes idle. As an example,
	 * this happens during idle_balance() which calls
	 * update_blocked_averages()
	 */
3216 3217
	if (!load)
		runnable = running = 0;
3218

3219 3220 3221 3222 3223 3224 3225
	/*
	 * Now we know we crossed measurement unit boundaries. The *_avg
	 * accrues by two steps:
	 *
	 * Step 1: accumulate *_sum since last_update_time. If we haven't
	 * crossed period boundaries, finish.
	 */
3226
	if (!accumulate_sum(delta, cpu, sa, load, runnable, running))
3227
		return 0;
3228

3229 3230 3231 3232
	return 1;
}

static __always_inline void
3233
___update_load_avg(struct sched_avg *sa, unsigned long load, unsigned long runnable)
3234 3235 3236
{
	u32 divider = LOAD_AVG_MAX - 1024 + sa->period_contrib;

3237 3238 3239
	/*
	 * Step 2: update *_avg.
	 */
3240 3241
	sa->load_avg = div_u64(load * sa->load_sum, divider);
	sa->runnable_load_avg =	div_u64(runnable * sa->runnable_load_sum, divider);
3242 3243
	sa->util_avg = sa->util_sum / divider;
}
3244

3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270
/*
 * When a task is dequeued, its estimated utilization should not be update if
 * its util_avg has not been updated at least once.
 * This flag is used to synchronize util_avg updates with util_est updates.
 * We map this information into the LSB bit of the utilization saved at
 * dequeue time (i.e. util_est.dequeued).
 */
#define UTIL_AVG_UNCHANGED 0x1

static inline void cfs_se_util_change(struct sched_avg *avg)
{
	unsigned int enqueued;

	if (!sched_feat(UTIL_EST))
		return;

	/* Avoid store if the flag has been already set */
	enqueued = avg->util_est.enqueued;
	if (!(enqueued & UTIL_AVG_UNCHANGED))
		return;

	/* Reset flag to report util_avg has been updated */
	enqueued &= ~UTIL_AVG_UNCHANGED;
	WRITE_ONCE(avg->util_est.enqueued, enqueued);
}

3271 3272 3273
/*
 * sched_entity:
 *
3274 3275 3276 3277 3278 3279 3280
 *   task:
 *     se_runnable() == se_weight()
 *
 *   group: [ see update_cfs_group() ]
 *     se_weight()   = tg->weight * grq->load_avg / tg->load_avg
 *     se_runnable() = se_weight(se) * grq->runnable_load_avg / grq->load_avg
 *
3281 3282 3283
 *   load_sum := runnable_sum
 *   load_avg = se_weight(se) * runnable_avg
 *
3284 3285 3286 3287 3288
 *   runnable_load_sum := runnable_sum
 *   runnable_load_avg = se_runnable(se) * runnable_avg
 *
 * XXX collapse load_sum and runnable_load_sum
 *
3289 3290 3291 3292
 * cfq_rs:
 *
 *   load_sum = \Sum se_weight(se) * se->avg.load_sum
 *   load_avg = \Sum se->avg.load_avg
3293 3294 3295
 *
 *   runnable_load_sum = \Sum se_runnable(se) * se->avg.runnable_load_sum
 *   runnable_load_avg = \Sum se->avg.runable_load_avg
3296 3297
 */

3298 3299 3300
static int
__update_load_avg_blocked_se(u64 now, int cpu, struct sched_entity *se)
{
3301 3302 3303 3304 3305
	if (entity_is_task(se))
		se->runnable_weight = se->load.weight;

	if (___update_load_sum(now, cpu, &se->avg, 0, 0, 0)) {
		___update_load_avg(&se->avg, se_weight(se), se_runnable(se));
3306 3307 3308 3309
		return 1;
	}

	return 0;
3310 3311 3312 3313 3314
}

static int
__update_load_avg_se(u64 now, int cpu, struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3315 3316 3317 3318 3319
	if (entity_is_task(se))
		se->runnable_weight = se->load.weight;

	if (___update_load_sum(now, cpu, &se->avg, !!se->on_rq, !!se->on_rq,
				cfs_rq->curr == se)) {
3320

3321
		___update_load_avg(&se->avg, se_weight(se), se_runnable(se));
3322
		cfs_se_util_change(&se->avg);
3323 3324 3325 3326
		return 1;
	}

	return 0;
3327 3328 3329 3330 3331
}

static int
__update_load_avg_cfs_rq(u64 now, int cpu, struct cfs_rq *cfs_rq)
{
3332 3333
	if (___update_load_sum(now, cpu, &cfs_rq->avg,
				scale_load_down(cfs_rq->load.weight),
3334 3335 3336 3337
				scale_load_down(cfs_rq->runnable_weight),
				cfs_rq->curr != NULL)) {

		___update_load_avg(&cfs_rq->avg, 1, 1);
3338 3339 3340 3341
		return 1;
	}

	return 0;
3342 3343
}

3344
#ifdef CONFIG_FAIR_GROUP_SCHED
3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357
/**
 * update_tg_load_avg - update the tg's load avg
 * @cfs_rq: the cfs_rq whose avg changed
 * @force: update regardless of how small the difference
 *
 * This function 'ensures': tg->load_avg := \Sum tg->cfs_rq[]->avg.load.
 * However, because tg->load_avg is a global value there are performance
 * considerations.
 *
 * In order to avoid having to look at the other cfs_rq's, we use a
 * differential update where we store the last value we propagated. This in
 * turn allows skipping updates if the differential is 'small'.
 *
3358
 * Updating tg's load_avg is necessary before update_cfs_share().
3359
 */
3360
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
3361
{
3362
	long delta = cfs_rq->avg.load_avg - cfs_rq->tg_load_avg_contrib;
3363

3364 3365 3366 3367 3368 3369
	/*
	 * No need to update load_avg for root_task_group as it is not used.
	 */
	if (cfs_rq->tg == &root_task_group)
		return;

3370 3371 3372
	if (force || abs(delta) > cfs_rq->tg_load_avg_contrib / 64) {
		atomic_long_add(delta, &cfs_rq->tg->load_avg);
		cfs_rq->tg_load_avg_contrib = cfs_rq->avg.load_avg;
3373
	}
3374
}
3375

3376
/*
3377
 * Called within set_task_rq() right before setting a task's CPU. The
3378 3379 3380 3381 3382 3383
 * caller only guarantees p->pi_lock is held; no other assumptions,
 * including the state of rq->lock, should be made.
 */
void set_task_rq_fair(struct sched_entity *se,
		      struct cfs_rq *prev, struct cfs_rq *next)
{
3384 3385 3386
	u64 p_last_update_time;
	u64 n_last_update_time;

3387 3388 3389 3390 3391 3392 3393 3394 3395 3396
	if (!sched_feat(ATTACH_AGE_LOAD))
		return;

	/*
	 * We are supposed to update the task to "current" time, then its up to
	 * date and ready to go to new CPU/cfs_rq. But we have difficulty in
	 * getting what current time is, so simply throw away the out-of-date
	 * time. This will result in the wakee task is less decayed, but giving
	 * the wakee more load sounds not bad.
	 */
3397 3398
	if (!(se->avg.last_update_time && prev))
		return;
3399 3400

#ifndef CONFIG_64BIT
3401
	{
3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415
		u64 p_last_update_time_copy;
		u64 n_last_update_time_copy;

		do {
			p_last_update_time_copy = prev->load_last_update_time_copy;
			n_last_update_time_copy = next->load_last_update_time_copy;

			smp_rmb();

			p_last_update_time = prev->avg.last_update_time;
			n_last_update_time = next->avg.last_update_time;

		} while (p_last_update_time != p_last_update_time_copy ||
			 n_last_update_time != n_last_update_time_copy);
3416
	}
3417
#else
3418 3419
	p_last_update_time = prev->avg.last_update_time;
	n_last_update_time = next->avg.last_update_time;
3420
#endif
3421 3422
	__update_load_avg_blocked_se(p_last_update_time, cpu_of(rq_of(prev)), se);
	se->avg.last_update_time = n_last_update_time;
3423
}
3424

3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435

/*
 * When on migration a sched_entity joins/leaves the PELT hierarchy, we need to
 * propagate its contribution. The key to this propagation is the invariant
 * that for each group:
 *
 *   ge->avg == grq->avg						(1)
 *
 * _IFF_ we look at the pure running and runnable sums. Because they
 * represent the very same entity, just at different points in the hierarchy.
 *
3436 3437 3438
 * Per the above update_tg_cfs_util() is trivial and simply copies the running
 * sum over (but still wrong, because the group entity and group rq do not have
 * their PELT windows aligned).
3439 3440 3441 3442 3443 3444 3445 3446
 *
 * However, update_tg_cfs_runnable() is more complex. So we have:
 *
 *   ge->avg.load_avg = ge->load.weight * ge->avg.runnable_avg		(2)
 *
 * And since, like util, the runnable part should be directly transferable,
 * the following would _appear_ to be the straight forward approach:
 *
3447
 *   grq->avg.load_avg = grq->load.weight * grq->avg.runnable_avg	(3)
3448 3449 3450
 *
 * And per (1) we have:
 *
3451
 *   ge->avg.runnable_avg == grq->avg.runnable_avg
3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469
 *
 * Which gives:
 *
 *                      ge->load.weight * grq->avg.load_avg
 *   ge->avg.load_avg = -----------------------------------		(4)
 *                               grq->load.weight
 *
 * Except that is wrong!
 *
 * Because while for entities historical weight is not important and we
 * really only care about our future and therefore can consider a pure
 * runnable sum, runqueues can NOT do this.
 *
 * We specifically want runqueues to have a load_avg that includes
 * historical weights. Those represent the blocked load, the load we expect
 * to (shortly) return to us. This only works by keeping the weights as
 * integral part of the sum. We therefore cannot decompose as per (3).
 *
3470 3471 3472 3473 3474 3475
 * Another reason this doesn't work is that runnable isn't a 0-sum entity.
 * Imagine a rq with 2 tasks that each are runnable 2/3 of the time. Then the
 * rq itself is runnable anywhere between 2/3 and 1 depending on how the
 * runnable section of these tasks overlap (or not). If they were to perfectly
 * align the rq as a whole would be runnable 2/3 of the time. If however we
 * always have at least 1 runnable task, the rq as a whole is always runnable.
3476
 *
3477
 * So we'll have to approximate.. :/
3478
 *
3479
 * Given the constraint:
3480
 *
3481
 *   ge->avg.running_sum <= ge->avg.runnable_sum <= LOAD_AVG_MAX
3482
 *
3483 3484
 * We can construct a rule that adds runnable to a rq by assuming minimal
 * overlap.
3485
 *
3486
 * On removal, we'll assume each task is equally runnable; which yields:
3487
 *
3488
 *   grq->avg.runnable_sum = grq->avg.load_sum / grq->load.weight
3489
 *
3490
 * XXX: only do this for the part of runnable > running ?
3491 3492 3493
 *
 */

3494
static inline void
3495
update_tg_cfs_util(struct cfs_rq *cfs_rq, struct sched_entity *se, struct cfs_rq *gcfs_rq)
3496 3497 3498 3499 3500 3501 3502
{
	long delta = gcfs_rq->avg.util_avg - se->avg.util_avg;

	/* Nothing to update */
	if (!delta)
		return;

3503 3504 3505 3506 3507 3508 3509 3510
	/*
	 * The relation between sum and avg is:
	 *
	 *   LOAD_AVG_MAX - 1024 + sa->period_contrib
	 *
	 * however, the PELT windows are not aligned between grq and gse.
	 */

3511 3512 3513 3514 3515 3516 3517 3518 3519 3520
	/* Set new sched_entity's utilization */
	se->avg.util_avg = gcfs_rq->avg.util_avg;
	se->avg.util_sum = se->avg.util_avg * LOAD_AVG_MAX;

	/* Update parent cfs_rq utilization */
	add_positive(&cfs_rq->avg.util_avg, delta);
	cfs_rq->avg.util_sum = cfs_rq->avg.util_avg * LOAD_AVG_MAX;
}

static inline void
3521
update_tg_cfs_runnable(struct cfs_rq *cfs_rq, struct sched_entity *se, struct cfs_rq *gcfs_rq)
3522
{
3523 3524 3525 3526
	long delta_avg, running_sum, runnable_sum = gcfs_rq->prop_runnable_sum;
	unsigned long runnable_load_avg, load_avg;
	u64 runnable_load_sum, load_sum = 0;
	s64 delta_sum;
3527

3528 3529
	if (!runnable_sum)
		return;
3530

3531
	gcfs_rq->prop_runnable_sum = 0;
3532

3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555
	if (runnable_sum >= 0) {
		/*
		 * Add runnable; clip at LOAD_AVG_MAX. Reflects that until
		 * the CPU is saturated running == runnable.
		 */
		runnable_sum += se->avg.load_sum;
		runnable_sum = min(runnable_sum, (long)LOAD_AVG_MAX);
	} else {
		/*
		 * Estimate the new unweighted runnable_sum of the gcfs_rq by
		 * assuming all tasks are equally runnable.
		 */
		if (scale_load_down(gcfs_rq->load.weight)) {
			load_sum = div_s64(gcfs_rq->avg.load_sum,
				scale_load_down(gcfs_rq->load.weight));
		}

		/* But make sure to not inflate se's runnable */
		runnable_sum = min(se->avg.load_sum, load_sum);
	}

	/*
	 * runnable_sum can't be lower than running_sum
3556
	 * As running sum is scale with CPU capacity wehreas the runnable sum
3557 3558 3559 3560 3561 3562
	 * is not we rescale running_sum 1st
	 */
	running_sum = se->avg.util_sum /
		arch_scale_cpu_capacity(NULL, cpu_of(rq_of(cfs_rq)));
	runnable_sum = max(runnable_sum, running_sum);

3563 3564
	load_sum = (s64)se_weight(se) * runnable_sum;
	load_avg = div_s64(load_sum, LOAD_AVG_MAX);
3565

3566 3567
	delta_sum = load_sum - (s64)se_weight(se) * se->avg.load_sum;
	delta_avg = load_avg - se->avg.load_avg;
3568

3569 3570 3571 3572
	se->avg.load_sum = runnable_sum;
	se->avg.load_avg = load_avg;
	add_positive(&cfs_rq->avg.load_avg, delta_avg);
	add_positive(&cfs_rq->avg.load_sum, delta_sum);
3573

3574 3575
	runnable_load_sum = (s64)se_runnable(se) * runnable_sum;
	runnable_load_avg = div_s64(runnable_load_sum, LOAD_AVG_MAX);
3576 3577
	delta_sum = runnable_load_sum - se_weight(se) * se->avg.runnable_load_sum;
	delta_avg = runnable_load_avg - se->avg.runnable_load_avg;
3578

3579 3580
	se->avg.runnable_load_sum = runnable_sum;
	se->avg.runnable_load_avg = runnable_load_avg;
3581

3582
	if (se->on_rq) {
3583 3584
		add_positive(&cfs_rq->avg.runnable_load_avg, delta_avg);
		add_positive(&cfs_rq->avg.runnable_load_sum, delta_sum);
3585 3586 3587
	}
}

3588
static inline void add_tg_cfs_propagate(struct cfs_rq *cfs_rq, long runnable_sum)
3589
{
3590 3591
	cfs_rq->propagate = 1;
	cfs_rq->prop_runnable_sum += runnable_sum;
3592 3593 3594 3595 3596
}

/* Update task and its cfs_rq load average */
static inline int propagate_entity_load_avg(struct sched_entity *se)
{
3597
	struct cfs_rq *cfs_rq, *gcfs_rq;
3598 3599 3600 3601

	if (entity_is_task(se))
		return 0;

3602 3603
	gcfs_rq = group_cfs_rq(se);
	if (!gcfs_rq->propagate)
3604 3605
		return 0;

3606 3607
	gcfs_rq->propagate = 0;

3608 3609
	cfs_rq = cfs_rq_of(se);

3610
	add_tg_cfs_propagate(cfs_rq, gcfs_rq->prop_runnable_sum);
3611

3612 3613
	update_tg_cfs_util(cfs_rq, se, gcfs_rq);
	update_tg_cfs_runnable(cfs_rq, se, gcfs_rq);
3614 3615 3616 3617

	return 1;
}

3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636
/*
 * Check if we need to update the load and the utilization of a blocked
 * group_entity:
 */
static inline bool skip_blocked_update(struct sched_entity *se)
{
	struct cfs_rq *gcfs_rq = group_cfs_rq(se);

	/*
	 * If sched_entity still have not zero load or utilization, we have to
	 * decay it:
	 */
	if (se->avg.load_avg || se->avg.util_avg)
		return false;

	/*
	 * If there is a pending propagation, we have to update the load and
	 * the utilization of the sched_entity:
	 */
3637
	if (gcfs_rq->propagate)
3638 3639 3640 3641 3642 3643 3644 3645 3646 3647
		return false;

	/*
	 * Otherwise, the load and the utilization of the sched_entity is
	 * already zero and there is no pending propagation, so it will be a
	 * waste of time to try to decay it:
	 */
	return true;
}

3648
#else /* CONFIG_FAIR_GROUP_SCHED */
3649

3650
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force) {}
3651 3652 3653 3654 3655 3656

static inline int propagate_entity_load_avg(struct sched_entity *se)
{
	return 0;
}

3657
static inline void add_tg_cfs_propagate(struct cfs_rq *cfs_rq, long runnable_sum) {}
3658

3659
#endif /* CONFIG_FAIR_GROUP_SCHED */
3660

3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671
/**
 * update_cfs_rq_load_avg - update the cfs_rq's load/util averages
 * @now: current time, as per cfs_rq_clock_task()
 * @cfs_rq: cfs_rq to update
 *
 * The cfs_rq avg is the direct sum of all its entities (blocked and runnable)
 * avg. The immediate corollary is that all (fair) tasks must be attached, see
 * post_init_entity_util_avg().
 *
 * cfs_rq->avg is used for task_h_load() and update_cfs_share() for example.
 *
3672 3673 3674 3675
 * Returns true if the load decayed or we removed load.
 *
 * Since both these conditions indicate a changed cfs_rq->avg.load we should
 * call update_tg_load_avg() when this function returns true.
3676
 */
3677
static inline int
3678
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq)
3679
{
3680
	unsigned long removed_load = 0, removed_util = 0, removed_runnable_sum = 0;
3681
	struct sched_avg *sa = &cfs_rq->avg;
3682
	int decayed = 0;
3683

3684 3685
	if (cfs_rq->removed.nr) {
		unsigned long r;
3686
		u32 divider = LOAD_AVG_MAX - 1024 + sa->period_contrib;
3687 3688 3689 3690

		raw_spin_lock(&cfs_rq->removed.lock);
		swap(cfs_rq->removed.util_avg, removed_util);
		swap(cfs_rq->removed.load_avg, removed_load);
3691
		swap(cfs_rq->removed.runnable_sum, removed_runnable_sum);
3692 3693 3694 3695
		cfs_rq->removed.nr = 0;
		raw_spin_unlock(&cfs_rq->removed.lock);

		r = removed_load;
3696
		sub_positive(&sa->load_avg, r);
3697
		sub_positive(&sa->load_sum, r * divider);
3698

3699
		r = removed_util;
3700
		sub_positive(&sa->util_avg, r);
3701
		sub_positive(&sa->util_sum, r * divider);
3702

3703
		add_tg_cfs_propagate(cfs_rq, -(long)removed_runnable_sum);
3704 3705

		decayed = 1;
3706
	}
3707

3708
	decayed |= __update_load_avg_cfs_rq(now, cpu_of(rq_of(cfs_rq)), cfs_rq);
3709

3710 3711 3712 3713
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->load_last_update_time_copy = sa->last_update_time;
#endif
3714

3715
	if (decayed)
3716
		cfs_rq_util_change(cfs_rq, 0);
3717

3718
	return decayed;
3719 3720
}

3721 3722 3723 3724 3725 3726 3727 3728
/**
 * attach_entity_load_avg - attach this entity to its cfs_rq load avg
 * @cfs_rq: cfs_rq to attach to
 * @se: sched_entity to attach
 *
 * Must call update_cfs_rq_load_avg() before this, since we rely on
 * cfs_rq->avg.last_update_time being current.
 */
3729
static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3730
{
3731 3732 3733 3734 3735 3736 3737 3738 3739
	u32 divider = LOAD_AVG_MAX - 1024 + cfs_rq->avg.period_contrib;

	/*
	 * When we attach the @se to the @cfs_rq, we must align the decay
	 * window because without that, really weird and wonderful things can
	 * happen.
	 *
	 * XXX illustrate
	 */
3740
	se->avg.last_update_time = cfs_rq->avg.last_update_time;
3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758
	se->avg.period_contrib = cfs_rq->avg.period_contrib;

	/*
	 * Hell(o) Nasty stuff.. we need to recompute _sum based on the new
	 * period_contrib. This isn't strictly correct, but since we're
	 * entirely outside of the PELT hierarchy, nobody cares if we truncate
	 * _sum a little.
	 */
	se->avg.util_sum = se->avg.util_avg * divider;

	se->avg.load_sum = divider;
	if (se_weight(se)) {
		se->avg.load_sum =
			div_u64(se->avg.load_avg * se->avg.load_sum, se_weight(se));
	}

	se->avg.runnable_load_sum = se->avg.load_sum;

3759
	enqueue_load_avg(cfs_rq, se);
3760 3761
	cfs_rq->avg.util_avg += se->avg.util_avg;
	cfs_rq->avg.util_sum += se->avg.util_sum;
3762 3763

	add_tg_cfs_propagate(cfs_rq, se->avg.load_sum);
3764

3765
	cfs_rq_util_change(cfs_rq, flags);
3766 3767
}

3768 3769 3770 3771 3772 3773 3774 3775
/**
 * detach_entity_load_avg - detach this entity from its cfs_rq load avg
 * @cfs_rq: cfs_rq to detach from
 * @se: sched_entity to detach
 *
 * Must call update_cfs_rq_load_avg() before this, since we rely on
 * cfs_rq->avg.last_update_time being current.
 */
3776 3777
static void detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3778
	dequeue_load_avg(cfs_rq, se);
3779 3780
	sub_positive(&cfs_rq->avg.util_avg, se->avg.util_avg);
	sub_positive(&cfs_rq->avg.util_sum, se->avg.util_sum);
3781 3782

	add_tg_cfs_propagate(cfs_rq, -se->avg.load_sum);
3783

3784
	cfs_rq_util_change(cfs_rq, 0);
3785 3786
}

3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813
/*
 * Optional action to be done while updating the load average
 */
#define UPDATE_TG	0x1
#define SKIP_AGE_LOAD	0x2
#define DO_ATTACH	0x4

/* Update task and its cfs_rq load average */
static inline void update_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
{
	u64 now = cfs_rq_clock_task(cfs_rq);
	struct rq *rq = rq_of(cfs_rq);
	int cpu = cpu_of(rq);
	int decayed;

	/*
	 * Track task load average for carrying it to new CPU after migrated, and
	 * track group sched_entity load average for task_h_load calc in migration
	 */
	if (se->avg.last_update_time && !(flags & SKIP_AGE_LOAD))
		__update_load_avg_se(now, cpu, cfs_rq, se);

	decayed  = update_cfs_rq_load_avg(now, cfs_rq);
	decayed |= propagate_entity_load_avg(se);

	if (!se->avg.last_update_time && (flags & DO_ATTACH)) {

3814 3815 3816 3817 3818 3819 3820 3821
		/*
		 * DO_ATTACH means we're here from enqueue_entity().
		 * !last_update_time means we've passed through
		 * migrate_task_rq_fair() indicating we migrated.
		 *
		 * IOW we're enqueueing a task on a new CPU.
		 */
		attach_entity_load_avg(cfs_rq, se, SCHED_CPUFREQ_MIGRATION);
3822 3823 3824 3825 3826 3827
		update_tg_load_avg(cfs_rq, 0);

	} else if (decayed && (flags & UPDATE_TG))
		update_tg_load_avg(cfs_rq, 0);
}

3828
#ifndef CONFIG_64BIT
3829 3830
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
3831
	u64 last_update_time_copy;
3832
	u64 last_update_time;
3833

3834 3835 3836 3837 3838
	do {
		last_update_time_copy = cfs_rq->load_last_update_time_copy;
		smp_rmb();
		last_update_time = cfs_rq->avg.last_update_time;
	} while (last_update_time != last_update_time_copy);
3839 3840 3841

	return last_update_time;
}
3842
#else
3843 3844 3845 3846
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
	return cfs_rq->avg.last_update_time;
}
3847 3848
#endif

3849 3850 3851 3852 3853 3854 3855 3856 3857 3858
/*
 * Synchronize entity load avg of dequeued entity without locking
 * the previous rq.
 */
void sync_entity_load_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 last_update_time;

	last_update_time = cfs_rq_last_update_time(cfs_rq);
3859
	__update_load_avg_blocked_se(last_update_time, cpu_of(rq_of(cfs_rq)), se);
3860 3861
}

3862 3863 3864 3865 3866 3867 3868
/*
 * Task first catches up with cfs_rq, and then subtract
 * itself from the cfs_rq (task must be off the queue now).
 */
void remove_entity_load_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
3869
	unsigned long flags;
3870 3871

	/*
3872 3873 3874 3875 3876 3877 3878
	 * tasks cannot exit without having gone through wake_up_new_task() ->
	 * post_init_entity_util_avg() which will have added things to the
	 * cfs_rq, so we can remove unconditionally.
	 *
	 * Similarly for groups, they will have passed through
	 * post_init_entity_util_avg() before unregister_sched_fair_group()
	 * calls this.
3879 3880
	 */

3881
	sync_entity_load_avg(se);
3882 3883 3884 3885 3886

	raw_spin_lock_irqsave(&cfs_rq->removed.lock, flags);
	++cfs_rq->removed.nr;
	cfs_rq->removed.util_avg	+= se->avg.util_avg;
	cfs_rq->removed.load_avg	+= se->avg.load_avg;
3887
	cfs_rq->removed.runnable_sum	+= se->avg.load_sum; /* == runnable_sum */
3888
	raw_spin_unlock_irqrestore(&cfs_rq->removed.lock, flags);
3889
}
3890

3891 3892
static inline unsigned long cfs_rq_runnable_load_avg(struct cfs_rq *cfs_rq)
{
3893
	return cfs_rq->avg.runnable_load_avg;
3894 3895 3896 3897 3898 3899 3900
}

static inline unsigned long cfs_rq_load_avg(struct cfs_rq *cfs_rq)
{
	return cfs_rq->avg.load_avg;
}

3901
static int idle_balance(struct rq *this_rq, struct rq_flags *rf);
3902

3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929
static inline unsigned long task_util(struct task_struct *p)
{
	return READ_ONCE(p->se.avg.util_avg);
}

static inline unsigned long _task_util_est(struct task_struct *p)
{
	struct util_est ue = READ_ONCE(p->se.avg.util_est);

	return max(ue.ewma, ue.enqueued);
}

static inline unsigned long task_util_est(struct task_struct *p)
{
	return max(task_util(p), _task_util_est(p));
}

static inline void util_est_enqueue(struct cfs_rq *cfs_rq,
				    struct task_struct *p)
{
	unsigned int enqueued;

	if (!sched_feat(UTIL_EST))
		return;

	/* Update root cfs_rq's estimated utilization */
	enqueued  = cfs_rq->avg.util_est.enqueued;
3930
	enqueued += (_task_util_est(p) | UTIL_AVG_UNCHANGED);
3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965
	WRITE_ONCE(cfs_rq->avg.util_est.enqueued, enqueued);
}

/*
 * Check if a (signed) value is within a specified (unsigned) margin,
 * based on the observation that:
 *
 *     abs(x) < y := (unsigned)(x + y - 1) < (2 * y - 1)
 *
 * NOTE: this only works when value + maring < INT_MAX.
 */
static inline bool within_margin(int value, int margin)
{
	return ((unsigned int)(value + margin - 1) < (2 * margin - 1));
}

static void
util_est_dequeue(struct cfs_rq *cfs_rq, struct task_struct *p, bool task_sleep)
{
	long last_ewma_diff;
	struct util_est ue;

	if (!sched_feat(UTIL_EST))
		return;

	/*
	 * Update root cfs_rq's estimated utilization
	 *
	 * If *p is the last task then the root cfs_rq's estimated utilization
	 * of a CPU is 0 by definition.
	 */
	ue.enqueued = 0;
	if (cfs_rq->nr_running) {
		ue.enqueued  = cfs_rq->avg.util_est.enqueued;
		ue.enqueued -= min_t(unsigned int, ue.enqueued,
3966
				     (_task_util_est(p) | UTIL_AVG_UNCHANGED));
3967 3968 3969 3970 3971 3972 3973 3974 3975 3976
	}
	WRITE_ONCE(cfs_rq->avg.util_est.enqueued, ue.enqueued);

	/*
	 * Skip update of task's estimated utilization when the task has not
	 * yet completed an activation, e.g. being migrated.
	 */
	if (!task_sleep)
		return;

3977 3978 3979 3980 3981 3982 3983 3984
	/*
	 * If the PELT values haven't changed since enqueue time,
	 * skip the util_est update.
	 */
	ue = p->se.avg.util_est;
	if (ue.enqueued & UTIL_AVG_UNCHANGED)
		return;

3985 3986 3987 3988
	/*
	 * Skip update of task's estimated utilization when its EWMA is
	 * already ~1% close to its last activation value.
	 */
3989
	ue.enqueued = (task_util(p) | UTIL_AVG_UNCHANGED);
3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016
	last_ewma_diff = ue.enqueued - ue.ewma;
	if (within_margin(last_ewma_diff, (SCHED_CAPACITY_SCALE / 100)))
		return;

	/*
	 * Update Task's estimated utilization
	 *
	 * When *p completes an activation we can consolidate another sample
	 * of the task size. This is done by storing the current PELT value
	 * as ue.enqueued and by using this value to update the Exponential
	 * Weighted Moving Average (EWMA):
	 *
	 *  ewma(t) = w *  task_util(p) + (1-w) * ewma(t-1)
	 *          = w *  task_util(p) +         ewma(t-1)  - w * ewma(t-1)
	 *          = w * (task_util(p) -         ewma(t-1)) +     ewma(t-1)
	 *          = w * (      last_ewma_diff            ) +     ewma(t-1)
	 *          = w * (last_ewma_diff  +  ewma(t-1) / w)
	 *
	 * Where 'w' is the weight of new samples, which is configured to be
	 * 0.25, thus making w=1/4 ( >>= UTIL_EST_WEIGHT_SHIFT)
	 */
	ue.ewma <<= UTIL_EST_WEIGHT_SHIFT;
	ue.ewma  += last_ewma_diff;
	ue.ewma >>= UTIL_EST_WEIGHT_SHIFT;
	WRITE_ONCE(p->se.avg.util_est, ue);
}

4017 4018
#else /* CONFIG_SMP */

4019
static inline int
4020
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq)
4021 4022 4023 4024
{
	return 0;
}

4025 4026
#define UPDATE_TG	0x0
#define SKIP_AGE_LOAD	0x0
4027
#define DO_ATTACH	0x0
4028

4029
static inline void update_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int not_used1)
4030
{
4031
	cfs_rq_util_change(cfs_rq, 0);
4032 4033
}

4034
static inline void remove_entity_load_avg(struct sched_entity *se) {}
4035

4036
static inline void
4037
attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) {}
4038 4039 4040
static inline void
detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}

4041
static inline int idle_balance(struct rq *rq, struct rq_flags *rf)
4042 4043 4044 4045
{
	return 0;
}

4046 4047 4048 4049 4050 4051 4052
static inline void
util_est_enqueue(struct cfs_rq *cfs_rq, struct task_struct *p) {}

static inline void
util_est_dequeue(struct cfs_rq *cfs_rq, struct task_struct *p,
		 bool task_sleep) {}

4053
#endif /* CONFIG_SMP */
4054

P
Peter Zijlstra 已提交
4055 4056 4057 4058 4059 4060 4061 4062 4063
static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
#ifdef CONFIG_SCHED_DEBUG
	s64 d = se->vruntime - cfs_rq->min_vruntime;

	if (d < 0)
		d = -d;

	if (d > 3*sysctl_sched_latency)
4064
		schedstat_inc(cfs_rq->nr_spread_over);
P
Peter Zijlstra 已提交
4065 4066 4067
#endif
}

4068 4069 4070
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
4071
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
4072

4073 4074 4075 4076 4077 4078
	/*
	 * The 'current' period is already promised to the current tasks,
	 * however the extra weight of the new task will slow them down a
	 * little, place the new task so that it fits in the slot that
	 * stays open at the end.
	 */
P
Peter Zijlstra 已提交
4079
	if (initial && sched_feat(START_DEBIT))
4080
		vruntime += sched_vslice(cfs_rq, se);
4081

4082
	/* sleeps up to a single latency don't count. */
4083
	if (!initial) {
4084
		unsigned long thresh = sysctl_sched_latency;
4085

4086 4087 4088 4089 4090 4091
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
4092

4093
		vruntime -= thresh;
4094 4095
	}

4096
	/* ensure we never gain time by being placed backwards. */
4097
	se->vruntime = max_vruntime(se->vruntime, vruntime);
4098 4099
}

4100 4101
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113
static inline void check_schedstat_required(void)
{
#ifdef CONFIG_SCHEDSTATS
	if (schedstat_enabled())
		return;

	/* Force schedstat enabled if a dependent tracepoint is active */
	if (trace_sched_stat_wait_enabled()    ||
			trace_sched_stat_sleep_enabled()   ||
			trace_sched_stat_iowait_enabled()  ||
			trace_sched_stat_blocked_enabled() ||
			trace_sched_stat_runtime_enabled())  {
4114
		printk_deferred_once("Scheduler tracepoints stat_sleep, stat_iowait, "
4115
			     "stat_blocked and stat_runtime require the "
4116
			     "kernel parameter schedstats=enable or "
4117 4118 4119 4120 4121
			     "kernel.sched_schedstats=1\n");
	}
#endif
}

4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140

/*
 * MIGRATION
 *
 *	dequeue
 *	  update_curr()
 *	    update_min_vruntime()
 *	  vruntime -= min_vruntime
 *
 *	enqueue
 *	  update_curr()
 *	    update_min_vruntime()
 *	  vruntime += min_vruntime
 *
 * this way the vruntime transition between RQs is done when both
 * min_vruntime are up-to-date.
 *
 * WAKEUP (remote)
 *
4141
 *	->migrate_task_rq_fair() (p->state == TASK_WAKING)
4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152
 *	  vruntime -= min_vruntime
 *
 *	enqueue
 *	  update_curr()
 *	    update_min_vruntime()
 *	  vruntime += min_vruntime
 *
 * this way we don't have the most up-to-date min_vruntime on the originating
 * CPU and an up-to-date min_vruntime on the destination CPU.
 */

4153
static void
4154
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
4155
{
4156 4157 4158
	bool renorm = !(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_MIGRATED);
	bool curr = cfs_rq->curr == se;

4159
	/*
4160 4161
	 * If we're the current task, we must renormalise before calling
	 * update_curr().
4162
	 */
4163
	if (renorm && curr)
4164 4165
		se->vruntime += cfs_rq->min_vruntime;

4166 4167
	update_curr(cfs_rq);

4168
	/*
4169 4170 4171 4172
	 * Otherwise, renormalise after, such that we're placed at the current
	 * moment in time, instead of some random moment in the past. Being
	 * placed in the past could significantly boost this task to the
	 * fairness detriment of existing tasks.
4173
	 */
4174 4175 4176
	if (renorm && !curr)
		se->vruntime += cfs_rq->min_vruntime;

4177 4178 4179 4180 4181 4182 4183 4184
	/*
	 * When enqueuing a sched_entity, we must:
	 *   - Update loads to have both entity and cfs_rq synced with now.
	 *   - Add its load to cfs_rq->runnable_avg
	 *   - For group_entity, update its weight to reflect the new share of
	 *     its group cfs_rq
	 *   - Add its new weight to cfs_rq->load.weight
	 */
4185
	update_load_avg(cfs_rq, se, UPDATE_TG | DO_ATTACH);
4186
	update_cfs_group(se);
4187
	enqueue_runnable_load_avg(cfs_rq, se);
4188
	account_entity_enqueue(cfs_rq, se);
4189

4190
	if (flags & ENQUEUE_WAKEUP)
4191
		place_entity(cfs_rq, se, 0);
4192

4193
	check_schedstat_required();
4194 4195
	update_stats_enqueue(cfs_rq, se, flags);
	check_spread(cfs_rq, se);
4196
	if (!curr)
4197
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
4198
	se->on_rq = 1;
4199

4200
	if (cfs_rq->nr_running == 1) {
4201
		list_add_leaf_cfs_rq(cfs_rq);
4202 4203
		check_enqueue_throttle(cfs_rq);
	}
4204 4205
}

4206
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
4207
{
4208 4209
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
4210
		if (cfs_rq->last != se)
4211
			break;
4212 4213

		cfs_rq->last = NULL;
4214 4215
	}
}
P
Peter Zijlstra 已提交
4216

4217 4218 4219 4220
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
4221
		if (cfs_rq->next != se)
4222
			break;
4223 4224

		cfs_rq->next = NULL;
4225
	}
P
Peter Zijlstra 已提交
4226 4227
}

4228 4229 4230 4231
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
4232
		if (cfs_rq->skip != se)
4233
			break;
4234 4235

		cfs_rq->skip = NULL;
4236 4237 4238
	}
}

P
Peter Zijlstra 已提交
4239 4240
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
4241 4242 4243 4244 4245
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
4246 4247 4248

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
4249 4250
}

4251
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
4252

4253
static void
4254
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
4255
{
4256 4257 4258 4259
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
4260 4261 4262 4263 4264 4265 4266 4267 4268

	/*
	 * When dequeuing a sched_entity, we must:
	 *   - Update loads to have both entity and cfs_rq synced with now.
	 *   - Substract its load from the cfs_rq->runnable_avg.
	 *   - Substract its previous weight from cfs_rq->load.weight.
	 *   - For group entity, update its weight to reflect the new share
	 *     of its group cfs_rq.
	 */
4269
	update_load_avg(cfs_rq, se, UPDATE_TG);
4270
	dequeue_runnable_load_avg(cfs_rq, se);
4271

4272
	update_stats_dequeue(cfs_rq, se, flags);
P
Peter Zijlstra 已提交
4273

P
Peter Zijlstra 已提交
4274
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
4275

4276
	if (se != cfs_rq->curr)
4277
		__dequeue_entity(cfs_rq, se);
4278
	se->on_rq = 0;
4279
	account_entity_dequeue(cfs_rq, se);
4280 4281

	/*
4282 4283 4284 4285
	 * Normalize after update_curr(); which will also have moved
	 * min_vruntime if @se is the one holding it back. But before doing
	 * update_min_vruntime() again, which will discount @se's position and
	 * can move min_vruntime forward still more.
4286
	 */
4287
	if (!(flags & DEQUEUE_SLEEP))
4288
		se->vruntime -= cfs_rq->min_vruntime;
4289

4290 4291 4292
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

4293
	update_cfs_group(se);
4294 4295 4296 4297 4298 4299 4300 4301 4302

	/*
	 * Now advance min_vruntime if @se was the entity holding it back,
	 * except when: DEQUEUE_SAVE && !DEQUEUE_MOVE, in this case we'll be
	 * put back on, and if we advance min_vruntime, we'll be placed back
	 * further than we started -- ie. we'll be penalized.
	 */
	if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) == DEQUEUE_SAVE)
		update_min_vruntime(cfs_rq);
4303 4304 4305 4306 4307
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
4308
static void
I
Ingo Molnar 已提交
4309
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
4310
{
4311
	unsigned long ideal_runtime, delta_exec;
4312 4313
	struct sched_entity *se;
	s64 delta;
4314

P
Peter Zijlstra 已提交
4315
	ideal_runtime = sched_slice(cfs_rq, curr);
4316
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
4317
	if (delta_exec > ideal_runtime) {
4318
		resched_curr(rq_of(cfs_rq));
4319 4320 4321 4322 4323
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334
		return;
	}

	/*
	 * Ensure that a task that missed wakeup preemption by a
	 * narrow margin doesn't have to wait for a full slice.
	 * This also mitigates buddy induced latencies under load.
	 */
	if (delta_exec < sysctl_sched_min_granularity)
		return;

4335 4336
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
4337

4338 4339
	if (delta < 0)
		return;
4340

4341
	if (delta > ideal_runtime)
4342
		resched_curr(rq_of(cfs_rq));
4343 4344
}

4345
static void
4346
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
4347
{
4348 4349 4350 4351 4352 4353 4354
	/* 'current' is not kept within the tree. */
	if (se->on_rq) {
		/*
		 * Any task has to be enqueued before it get to execute on
		 * a CPU. So account for the time it spent waiting on the
		 * runqueue.
		 */
4355
		update_stats_wait_end(cfs_rq, se);
4356
		__dequeue_entity(cfs_rq, se);
4357
		update_load_avg(cfs_rq, se, UPDATE_TG);
4358 4359
	}

4360
	update_stats_curr_start(cfs_rq, se);
4361
	cfs_rq->curr = se;
4362

I
Ingo Molnar 已提交
4363 4364 4365 4366 4367
	/*
	 * Track our maximum slice length, if the CPU's load is at
	 * least twice that of our own weight (i.e. dont track it
	 * when there are only lesser-weight tasks around):
	 */
4368
	if (schedstat_enabled() && rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
4369 4370 4371
		schedstat_set(se->statistics.slice_max,
			max((u64)schedstat_val(se->statistics.slice_max),
			    se->sum_exec_runtime - se->prev_sum_exec_runtime));
I
Ingo Molnar 已提交
4372
	}
4373

4374
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
4375 4376
}

4377 4378 4379
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

4380 4381 4382 4383 4384 4385 4386
/*
 * Pick the next process, keeping these things in mind, in this order:
 * 1) keep things fair between processes/task groups
 * 2) pick the "next" process, since someone really wants that to run
 * 3) pick the "last" process, for cache locality
 * 4) do not run the "skip" process, if something else is available
 */
4387 4388
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
4389
{
4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400
	struct sched_entity *left = __pick_first_entity(cfs_rq);
	struct sched_entity *se;

	/*
	 * If curr is set we have to see if its left of the leftmost entity
	 * still in the tree, provided there was anything in the tree at all.
	 */
	if (!left || (curr && entity_before(curr, left)))
		left = curr;

	se = left; /* ideally we run the leftmost entity */
4401

4402 4403 4404 4405 4406
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
4407 4408 4409 4410 4411 4412 4413 4414 4415 4416
		struct sched_entity *second;

		if (se == curr) {
			second = __pick_first_entity(cfs_rq);
		} else {
			second = __pick_next_entity(se);
			if (!second || (curr && entity_before(curr, second)))
				second = curr;
		}

4417 4418 4419
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
4420

4421 4422 4423 4424 4425 4426
	/*
	 * Prefer last buddy, try to return the CPU to a preempted task.
	 */
	if (cfs_rq->last && wakeup_preempt_entity(cfs_rq->last, left) < 1)
		se = cfs_rq->last;

4427 4428 4429 4430 4431 4432
	/*
	 * Someone really wants this to run. If it's not unfair, run it.
	 */
	if (cfs_rq->next && wakeup_preempt_entity(cfs_rq->next, left) < 1)
		se = cfs_rq->next;

4433
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
4434 4435

	return se;
4436 4437
}

4438
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
4439

4440
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
4441 4442 4443 4444 4445 4446
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
4447
		update_curr(cfs_rq);
4448

4449 4450 4451
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

4452
	check_spread(cfs_rq, prev);
4453

4454
	if (prev->on_rq) {
4455
		update_stats_wait_start(cfs_rq, prev);
4456 4457
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
4458
		/* in !on_rq case, update occurred at dequeue */
4459
		update_load_avg(cfs_rq, prev, 0);
4460
	}
4461
	cfs_rq->curr = NULL;
4462 4463
}

P
Peter Zijlstra 已提交
4464 4465
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
4466 4467
{
	/*
4468
	 * Update run-time statistics of the 'current'.
4469
	 */
4470
	update_curr(cfs_rq);
4471

4472 4473 4474
	/*
	 * Ensure that runnable average is periodically updated.
	 */
4475
	update_load_avg(cfs_rq, curr, UPDATE_TG);
4476
	update_cfs_group(curr);
4477

P
Peter Zijlstra 已提交
4478 4479 4480 4481 4482
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
4483
	if (queued) {
4484
		resched_curr(rq_of(cfs_rq));
4485 4486
		return;
	}
P
Peter Zijlstra 已提交
4487 4488 4489 4490 4491 4492 4493 4494
	/*
	 * don't let the period tick interfere with the hrtick preemption
	 */
	if (!sched_feat(DOUBLE_TICK) &&
			hrtimer_active(&rq_of(cfs_rq)->hrtick_timer))
		return;
#endif

Y
Yong Zhang 已提交
4495
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
4496
		check_preempt_tick(cfs_rq, curr);
4497 4498
}

4499 4500 4501 4502 4503 4504

/**************************************************
 * CFS bandwidth control machinery
 */

#ifdef CONFIG_CFS_BANDWIDTH
4505 4506

#ifdef HAVE_JUMP_LABEL
4507
static struct static_key __cfs_bandwidth_used;
4508 4509 4510

static inline bool cfs_bandwidth_used(void)
{
4511
	return static_key_false(&__cfs_bandwidth_used);
4512 4513
}

4514
void cfs_bandwidth_usage_inc(void)
4515
{
4516
	static_key_slow_inc_cpuslocked(&__cfs_bandwidth_used);
4517 4518 4519 4520
}

void cfs_bandwidth_usage_dec(void)
{
4521
	static_key_slow_dec_cpuslocked(&__cfs_bandwidth_used);
4522 4523 4524 4525 4526 4527 4528
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

4529 4530
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
4531 4532
#endif /* HAVE_JUMP_LABEL */

4533 4534 4535 4536 4537 4538 4539 4540
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
4541 4542 4543 4544 4545 4546

static inline u64 sched_cfs_bandwidth_slice(void)
{
	return (u64)sysctl_sched_cfs_bandwidth_slice * NSEC_PER_USEC;
}

P
Paul Turner 已提交
4547 4548 4549 4550 4551 4552 4553
/*
 * Replenish runtime according to assigned quota and update expiration time.
 * We use sched_clock_cpu directly instead of rq->clock to avoid adding
 * additional synchronization around rq->lock.
 *
 * requires cfs_b->lock
 */
4554
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565
{
	u64 now;

	if (cfs_b->quota == RUNTIME_INF)
		return;

	now = sched_clock_cpu(smp_processor_id());
	cfs_b->runtime = cfs_b->quota;
	cfs_b->runtime_expires = now + ktime_to_ns(cfs_b->period);
}

4566 4567 4568 4569 4570
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

4571 4572 4573 4574
/* rq->task_clock normalized against any time this cfs_rq has spent throttled */
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
	if (unlikely(cfs_rq->throttle_count))
4575
		return cfs_rq->throttled_clock_task - cfs_rq->throttled_clock_task_time;
4576

4577
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
4578 4579
}

4580 4581
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4582 4583 4584
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
4585
	u64 amount = 0, min_amount, expires;
4586 4587 4588 4589 4590 4591 4592

	/* note: this is a positive sum as runtime_remaining <= 0 */
	min_amount = sched_cfs_bandwidth_slice() - cfs_rq->runtime_remaining;

	raw_spin_lock(&cfs_b->lock);
	if (cfs_b->quota == RUNTIME_INF)
		amount = min_amount;
4593
	else {
P
Peter Zijlstra 已提交
4594
		start_cfs_bandwidth(cfs_b);
4595 4596 4597 4598 4599 4600

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
4601
	}
P
Paul Turner 已提交
4602
	expires = cfs_b->runtime_expires;
4603 4604 4605
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
4606 4607 4608 4609 4610 4611 4612
	/*
	 * we may have advanced our local expiration to account for allowed
	 * spread between our sched_clock and the one on which runtime was
	 * issued.
	 */
	if ((s64)(expires - cfs_rq->runtime_expires) > 0)
		cfs_rq->runtime_expires = expires;
4613 4614

	return cfs_rq->runtime_remaining > 0;
4615 4616
}

P
Paul Turner 已提交
4617 4618 4619 4620 4621
/*
 * Note: This depends on the synchronization provided by sched_clock and the
 * fact that rq->clock snapshots this value.
 */
static void expire_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4622
{
P
Paul Turner 已提交
4623 4624 4625
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
4626
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
4627 4628
		return;

P
Paul Turner 已提交
4629 4630 4631 4632 4633 4634 4635 4636 4637
	if (cfs_rq->runtime_remaining < 0)
		return;

	/*
	 * If the local deadline has passed we have to consider the
	 * possibility that our sched_clock is 'fast' and the global deadline
	 * has not truly expired.
	 *
	 * Fortunately we can check determine whether this the case by checking
4638 4639 4640
	 * whether the global deadline has advanced. It is valid to compare
	 * cfs_b->runtime_expires without any locks since we only care about
	 * exact equality, so a partial write will still work.
P
Paul Turner 已提交
4641 4642
	 */

4643
	if (cfs_rq->runtime_expires != cfs_b->runtime_expires) {
P
Paul Turner 已提交
4644 4645 4646 4647 4648 4649 4650 4651
		/* extend local deadline, drift is bounded above by 2 ticks */
		cfs_rq->runtime_expires += TICK_NSEC;
	} else {
		/* global deadline is ahead, expiration has passed */
		cfs_rq->runtime_remaining = 0;
	}
}

4652
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
4653 4654
{
	/* dock delta_exec before expiring quota (as it could span periods) */
4655
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
4656 4657 4658
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
4659 4660
		return;

4661 4662 4663 4664 4665
	/*
	 * if we're unable to extend our runtime we resched so that the active
	 * hierarchy can be throttled
	 */
	if (!assign_cfs_rq_runtime(cfs_rq) && likely(cfs_rq->curr))
4666
		resched_curr(rq_of(cfs_rq));
4667 4668
}

4669
static __always_inline
4670
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
4671
{
4672
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
4673 4674 4675 4676 4677
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

4678 4679
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
4680
	return cfs_bandwidth_used() && cfs_rq->throttled;
4681 4682
}

4683 4684 4685
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
4686
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713
}

/*
 * Ensure that neither of the group entities corresponding to src_cpu or
 * dest_cpu are members of a throttled hierarchy when performing group
 * load-balance operations.
 */
static inline int throttled_lb_pair(struct task_group *tg,
				    int src_cpu, int dest_cpu)
{
	struct cfs_rq *src_cfs_rq, *dest_cfs_rq;

	src_cfs_rq = tg->cfs_rq[src_cpu];
	dest_cfs_rq = tg->cfs_rq[dest_cpu];

	return throttled_hierarchy(src_cfs_rq) ||
	       throttled_hierarchy(dest_cfs_rq);
}

/* updated child weight may affect parent so we have to do this bottom up */
static int tg_unthrottle_up(struct task_group *tg, void *data)
{
	struct rq *rq = data;
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];

	cfs_rq->throttle_count--;
	if (!cfs_rq->throttle_count) {
4714
		/* adjust cfs_rq_clock_task() */
4715
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
4716
					     cfs_rq->throttled_clock_task;
4717 4718 4719 4720 4721 4722 4723 4724 4725 4726
	}

	return 0;
}

static int tg_throttle_down(struct task_group *tg, void *data)
{
	struct rq *rq = data;
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];

4727 4728
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
4729
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
4730 4731 4732 4733 4734
	cfs_rq->throttle_count++;

	return 0;
}

4735
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
4736 4737 4738 4739 4740
{
	struct rq *rq = rq_of(cfs_rq);
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
	struct sched_entity *se;
	long task_delta, dequeue = 1;
P
Peter Zijlstra 已提交
4741
	bool empty;
4742 4743 4744

	se = cfs_rq->tg->se[cpu_of(rq_of(cfs_rq))];

4745
	/* freeze hierarchy runnable averages while throttled */
4746 4747 4748
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765

	task_delta = cfs_rq->h_nr_running;
	for_each_sched_entity(se) {
		struct cfs_rq *qcfs_rq = cfs_rq_of(se);
		/* throttled entity or throttle-on-deactivate */
		if (!se->on_rq)
			break;

		if (dequeue)
			dequeue_entity(qcfs_rq, se, DEQUEUE_SLEEP);
		qcfs_rq->h_nr_running -= task_delta;

		if (qcfs_rq->load.weight)
			dequeue = 0;
	}

	if (!se)
4766
		sub_nr_running(rq, task_delta);
4767 4768

	cfs_rq->throttled = 1;
4769
	cfs_rq->throttled_clock = rq_clock(rq);
4770
	raw_spin_lock(&cfs_b->lock);
4771
	empty = list_empty(&cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
4772

4773 4774 4775 4776 4777
	/*
	 * Add to the _head_ of the list, so that an already-started
	 * distribute_cfs_runtime will not see us
	 */
	list_add_rcu(&cfs_rq->throttled_list, &cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
4778 4779 4780 4781 4782 4783 4784 4785

	/*
	 * If we're the first throttled task, make sure the bandwidth
	 * timer is running.
	 */
	if (empty)
		start_cfs_bandwidth(cfs_b);

4786 4787 4788
	raw_spin_unlock(&cfs_b->lock);
}

4789
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
4790 4791 4792 4793 4794 4795 4796
{
	struct rq *rq = rq_of(cfs_rq);
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
	struct sched_entity *se;
	int enqueue = 1;
	long task_delta;

4797
	se = cfs_rq->tg->se[cpu_of(rq)];
4798 4799

	cfs_rq->throttled = 0;
4800 4801 4802

	update_rq_clock(rq);

4803
	raw_spin_lock(&cfs_b->lock);
4804
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
4805 4806 4807
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

4808 4809 4810
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828
	if (!cfs_rq->load.weight)
		return;

	task_delta = cfs_rq->h_nr_running;
	for_each_sched_entity(se) {
		if (se->on_rq)
			enqueue = 0;

		cfs_rq = cfs_rq_of(se);
		if (enqueue)
			enqueue_entity(cfs_rq, se, ENQUEUE_WAKEUP);
		cfs_rq->h_nr_running += task_delta;

		if (cfs_rq_throttled(cfs_rq))
			break;
	}

	if (!se)
4829
		add_nr_running(rq, task_delta);
4830

4831
	/* Determine whether we need to wake up potentially idle CPU: */
4832
	if (rq->curr == rq->idle && rq->cfs.nr_running)
4833
		resched_curr(rq);
4834 4835 4836 4837 4838 4839
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
4840 4841
	u64 runtime;
	u64 starting_runtime = remaining;
4842 4843 4844 4845 4846

	rcu_read_lock();
	list_for_each_entry_rcu(cfs_rq, &cfs_b->throttled_cfs_rq,
				throttled_list) {
		struct rq *rq = rq_of(cfs_rq);
4847
		struct rq_flags rf;
4848

4849
		rq_lock(rq, &rf);
4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865
		if (!cfs_rq_throttled(cfs_rq))
			goto next;

		runtime = -cfs_rq->runtime_remaining + 1;
		if (runtime > remaining)
			runtime = remaining;
		remaining -= runtime;

		cfs_rq->runtime_remaining += runtime;
		cfs_rq->runtime_expires = expires;

		/* we check whether we're throttled above */
		if (cfs_rq->runtime_remaining > 0)
			unthrottle_cfs_rq(cfs_rq);

next:
4866
		rq_unlock(rq, &rf);
4867 4868 4869 4870 4871 4872

		if (!remaining)
			break;
	}
	rcu_read_unlock();

4873
	return starting_runtime - remaining;
4874 4875
}

4876 4877 4878 4879 4880 4881 4882 4883
/*
 * Responsible for refilling a task_group's bandwidth and unthrottling its
 * cfs_rqs as appropriate. If there has been no activity within the last
 * period the timer is deactivated until scheduling resumes; cfs_b->idle is
 * used to track this state.
 */
static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun)
{
4884
	u64 runtime, runtime_expires;
4885
	int throttled;
4886 4887 4888

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
4889
		goto out_deactivate;
4890

4891
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
4892
	cfs_b->nr_periods += overrun;
4893

4894 4895 4896 4897 4898 4899
	/*
	 * idle depends on !throttled (for the case of a large deficit), and if
	 * we're going inactive then everything else can be deferred
	 */
	if (cfs_b->idle && !throttled)
		goto out_deactivate;
P
Paul Turner 已提交
4900 4901 4902

	__refill_cfs_bandwidth_runtime(cfs_b);

4903 4904 4905
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
4906
		return 0;
4907 4908
	}

4909 4910 4911
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

4912 4913 4914
	runtime_expires = cfs_b->runtime_expires;

	/*
4915 4916 4917 4918 4919
	 * This check is repeated as we are holding onto the new bandwidth while
	 * we unthrottle. This can potentially race with an unthrottled group
	 * trying to acquire new bandwidth from the global pool. This can result
	 * in us over-using our runtime if it is all used during this loop, but
	 * only by limited amounts in that extreme case.
4920
	 */
4921 4922
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
4923 4924 4925 4926 4927 4928 4929
		raw_spin_unlock(&cfs_b->lock);
		/* we can't nest cfs_b->lock while distributing bandwidth */
		runtime = distribute_cfs_runtime(cfs_b, runtime,
						 runtime_expires);
		raw_spin_lock(&cfs_b->lock);

		throttled = !list_empty(&cfs_b->throttled_cfs_rq);
4930 4931

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4932
	}
4933

4934 4935 4936 4937 4938 4939 4940
	/*
	 * While we are ensured activity in the period following an
	 * unthrottle, this also covers the case in which the new bandwidth is
	 * insufficient to cover the existing bandwidth deficit.  (Forcing the
	 * timer to remain active while there are any throttled entities.)
	 */
	cfs_b->idle = 0;
4941

4942 4943 4944 4945
	return 0;

out_deactivate:
	return 1;
4946
}
4947

4948 4949 4950 4951 4952 4953 4954
/* a cfs_rq won't donate quota below this amount */
static const u64 min_cfs_rq_runtime = 1 * NSEC_PER_MSEC;
/* minimum remaining period time to redistribute slack quota */
static const u64 min_bandwidth_expiration = 2 * NSEC_PER_MSEC;
/* how long we wait to gather additional slack before distributing */
static const u64 cfs_bandwidth_slack_period = 5 * NSEC_PER_MSEC;

4955 4956 4957 4958
/*
 * Are we near the end of the current quota period?
 *
 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
4959
 * hrtimer base being cleared by hrtimer_start. In the case of
4960 4961
 * migrate_hrtimers, base is never cleared, so we are fine.
 */
4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986
static int runtime_refresh_within(struct cfs_bandwidth *cfs_b, u64 min_expire)
{
	struct hrtimer *refresh_timer = &cfs_b->period_timer;
	u64 remaining;

	/* if the call-back is running a quota refresh is already occurring */
	if (hrtimer_callback_running(refresh_timer))
		return 1;

	/* is a quota refresh about to occur? */
	remaining = ktime_to_ns(hrtimer_expires_remaining(refresh_timer));
	if (remaining < min_expire)
		return 1;

	return 0;
}

static void start_cfs_slack_bandwidth(struct cfs_bandwidth *cfs_b)
{
	u64 min_left = cfs_bandwidth_slack_period + min_bandwidth_expiration;

	/* if there's a quota refresh soon don't bother with slack */
	if (runtime_refresh_within(cfs_b, min_left))
		return;

P
Peter Zijlstra 已提交
4987 4988 4989
	hrtimer_start(&cfs_b->slack_timer,
			ns_to_ktime(cfs_bandwidth_slack_period),
			HRTIMER_MODE_REL);
4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018
}

/* we know any runtime found here is valid as update_curr() precedes return */
static void __return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
{
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
	s64 slack_runtime = cfs_rq->runtime_remaining - min_cfs_rq_runtime;

	if (slack_runtime <= 0)
		return;

	raw_spin_lock(&cfs_b->lock);
	if (cfs_b->quota != RUNTIME_INF &&
	    cfs_rq->runtime_expires == cfs_b->runtime_expires) {
		cfs_b->runtime += slack_runtime;

		/* we are under rq->lock, defer unthrottling using a timer */
		if (cfs_b->runtime > sched_cfs_bandwidth_slice() &&
		    !list_empty(&cfs_b->throttled_cfs_rq))
			start_cfs_slack_bandwidth(cfs_b);
	}
	raw_spin_unlock(&cfs_b->lock);

	/* even if it's not valid for return we don't want to try again */
	cfs_rq->runtime_remaining -= slack_runtime;
}

static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
{
5019 5020 5021
	if (!cfs_bandwidth_used())
		return;

5022
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037
		return;

	__return_cfs_rq_runtime(cfs_rq);
}

/*
 * This is done with a timer (instead of inline with bandwidth return) since
 * it's necessary to juggle rq->locks to unthrottle their respective cfs_rqs.
 */
static void do_sched_cfs_slack_timer(struct cfs_bandwidth *cfs_b)
{
	u64 runtime = 0, slice = sched_cfs_bandwidth_slice();
	u64 expires;

	/* confirm we're still not at a refresh boundary */
5038 5039 5040
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
5041
		return;
5042
	}
5043

5044
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
5045
		runtime = cfs_b->runtime;
5046

5047 5048 5049 5050 5051 5052 5053 5054 5055 5056
	expires = cfs_b->runtime_expires;
	raw_spin_unlock(&cfs_b->lock);

	if (!runtime)
		return;

	runtime = distribute_cfs_runtime(cfs_b, runtime, expires);

	raw_spin_lock(&cfs_b->lock);
	if (expires == cfs_b->runtime_expires)
5057
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
5058 5059 5060
	raw_spin_unlock(&cfs_b->lock);
}

5061 5062 5063 5064 5065 5066 5067
/*
 * When a group wakes up we want to make sure that its quota is not already
 * expired/exceeded, otherwise it may be allowed to steal additional ticks of
 * runtime as update_curr() throttling can not not trigger until it's on-rq.
 */
static void check_enqueue_throttle(struct cfs_rq *cfs_rq)
{
5068 5069 5070
	if (!cfs_bandwidth_used())
		return;

5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084
	/* an active group must be handled by the update_curr()->put() path */
	if (!cfs_rq->runtime_enabled || cfs_rq->curr)
		return;

	/* ensure the group is not already throttled */
	if (cfs_rq_throttled(cfs_rq))
		return;

	/* update runtime allocation */
	account_cfs_rq_runtime(cfs_rq, 0);
	if (cfs_rq->runtime_remaining <= 0)
		throttle_cfs_rq(cfs_rq);
}

5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098
static void sync_throttle(struct task_group *tg, int cpu)
{
	struct cfs_rq *pcfs_rq, *cfs_rq;

	if (!cfs_bandwidth_used())
		return;

	if (!tg->parent)
		return;

	cfs_rq = tg->cfs_rq[cpu];
	pcfs_rq = tg->parent->cfs_rq[cpu];

	cfs_rq->throttle_count = pcfs_rq->throttle_count;
5099
	cfs_rq->throttled_clock_task = rq_clock_task(cpu_rq(cpu));
5100 5101
}

5102
/* conditionally throttle active cfs_rq's from put_prev_entity() */
5103
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
5104
{
5105
	if (!cfs_bandwidth_used())
5106
		return false;
5107

5108
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
5109
		return false;
5110 5111 5112 5113 5114 5115

	/*
	 * it's possible for a throttled entity to be forced into a running
	 * state (e.g. set_curr_task), in this case we're finished.
	 */
	if (cfs_rq_throttled(cfs_rq))
5116
		return true;
5117 5118

	throttle_cfs_rq(cfs_rq);
5119
	return true;
5120
}
5121 5122 5123 5124 5125

static enum hrtimer_restart sched_cfs_slack_timer(struct hrtimer *timer)
{
	struct cfs_bandwidth *cfs_b =
		container_of(timer, struct cfs_bandwidth, slack_timer);
P
Peter Zijlstra 已提交
5126

5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138
	do_sched_cfs_slack_timer(cfs_b);

	return HRTIMER_NORESTART;
}

static enum hrtimer_restart sched_cfs_period_timer(struct hrtimer *timer)
{
	struct cfs_bandwidth *cfs_b =
		container_of(timer, struct cfs_bandwidth, period_timer);
	int overrun;
	int idle = 0;

5139
	raw_spin_lock(&cfs_b->lock);
5140
	for (;;) {
P
Peter Zijlstra 已提交
5141
		overrun = hrtimer_forward_now(timer, cfs_b->period);
5142 5143 5144 5145 5146
		if (!overrun)
			break;

		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}
P
Peter Zijlstra 已提交
5147 5148
	if (idle)
		cfs_b->period_active = 0;
5149
	raw_spin_unlock(&cfs_b->lock);
5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161

	return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
}

void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
	raw_spin_lock_init(&cfs_b->lock);
	cfs_b->runtime = 0;
	cfs_b->quota = RUNTIME_INF;
	cfs_b->period = ns_to_ktime(default_cfs_period());

	INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
5162
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173
	cfs_b->period_timer.function = sched_cfs_period_timer;
	hrtimer_init(&cfs_b->slack_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	cfs_b->slack_timer.function = sched_cfs_slack_timer;
}

static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq)
{
	cfs_rq->runtime_enabled = 0;
	INIT_LIST_HEAD(&cfs_rq->throttled_list);
}

P
Peter Zijlstra 已提交
5174
void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
5175
{
P
Peter Zijlstra 已提交
5176
	lockdep_assert_held(&cfs_b->lock);
5177

P
Peter Zijlstra 已提交
5178 5179 5180 5181 5182
	if (!cfs_b->period_active) {
		cfs_b->period_active = 1;
		hrtimer_forward_now(&cfs_b->period_timer, cfs_b->period);
		hrtimer_start_expires(&cfs_b->period_timer, HRTIMER_MODE_ABS_PINNED);
	}
5183 5184 5185 5186
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
5187 5188 5189 5190
	/* init_cfs_bandwidth() was not called */
	if (!cfs_b->throttled_cfs_rq.next)
		return;

5191 5192 5193 5194
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

5195
/*
5196
 * Both these CPU hotplug callbacks race against unregister_fair_sched_group()
5197 5198 5199 5200 5201 5202
 *
 * The race is harmless, since modifying bandwidth settings of unhooked group
 * bits doesn't do much.
 */

/* cpu online calback */
5203 5204
static void __maybe_unused update_runtime_enabled(struct rq *rq)
{
5205
	struct task_group *tg;
5206

5207 5208 5209 5210 5211 5212
	lockdep_assert_held(&rq->lock);

	rcu_read_lock();
	list_for_each_entry_rcu(tg, &task_groups, list) {
		struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
		struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
5213 5214 5215 5216 5217

		raw_spin_lock(&cfs_b->lock);
		cfs_rq->runtime_enabled = cfs_b->quota != RUNTIME_INF;
		raw_spin_unlock(&cfs_b->lock);
	}
5218
	rcu_read_unlock();
5219 5220
}

5221
/* cpu offline callback */
5222
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
5223
{
5224 5225 5226 5227 5228 5229 5230
	struct task_group *tg;

	lockdep_assert_held(&rq->lock);

	rcu_read_lock();
	list_for_each_entry_rcu(tg, &task_groups, list) {
		struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
5231 5232 5233 5234 5235 5236 5237 5238

		if (!cfs_rq->runtime_enabled)
			continue;

		/*
		 * clock_task is not advancing so we just need to make sure
		 * there's some valid quota amount
		 */
5239
		cfs_rq->runtime_remaining = 1;
5240
		/*
5241
		 * Offline rq is schedulable till CPU is completely disabled
5242 5243 5244 5245
		 * in take_cpu_down(), so we prevent new cfs throttling here.
		 */
		cfs_rq->runtime_enabled = 0;

5246 5247 5248
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
5249
	rcu_read_unlock();
5250 5251 5252
}

#else /* CONFIG_CFS_BANDWIDTH */
5253 5254
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
5255
	return rq_clock_task(rq_of(cfs_rq));
5256 5257
}

5258
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
5259
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
5260
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
5261
static inline void sync_throttle(struct task_group *tg, int cpu) {}
5262
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
5263 5264 5265 5266 5267

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278

static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
	return 0;
}

static inline int throttled_lb_pair(struct task_group *tg,
				    int src_cpu, int dest_cpu)
{
	return 0;
}
5279 5280 5281 5282 5283

void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {}

#ifdef CONFIG_FAIR_GROUP_SCHED
static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
5284 5285
#endif

5286 5287 5288 5289 5290
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return NULL;
}
static inline void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {}
5291
static inline void update_runtime_enabled(struct rq *rq) {}
5292
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
5293 5294 5295

#endif /* CONFIG_CFS_BANDWIDTH */

5296 5297 5298 5299
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
5300 5301 5302 5303 5304 5305
#ifdef CONFIG_SCHED_HRTICK
static void hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

5306
	SCHED_WARN_ON(task_rq(p) != rq);
P
Peter Zijlstra 已提交
5307

5308
	if (rq->cfs.h_nr_running > 1) {
P
Peter Zijlstra 已提交
5309 5310 5311 5312 5313 5314
		u64 slice = sched_slice(cfs_rq, se);
		u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime;
		s64 delta = slice - ran;

		if (delta < 0) {
			if (rq->curr == p)
5315
				resched_curr(rq);
P
Peter Zijlstra 已提交
5316 5317
			return;
		}
5318
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
5319 5320
	}
}
5321 5322 5323 5324 5325 5326 5327 5328 5329 5330

/*
 * called from enqueue/dequeue and updates the hrtick when the
 * current task is from our class and nr_running is low enough
 * to matter.
 */
static void hrtick_update(struct rq *rq)
{
	struct task_struct *curr = rq->curr;

5331
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
5332 5333 5334 5335 5336
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
5337
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
5338 5339 5340 5341
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
5342 5343 5344 5345

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
5346 5347
#endif

5348 5349 5350 5351 5352
/*
 * The enqueue_task method is called before nr_running is
 * increased. Here we update the fair scheduling stats and
 * then put the task into the rbtree:
 */
5353
static void
5354
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
5355 5356
{
	struct cfs_rq *cfs_rq;
5357
	struct sched_entity *se = &p->se;
5358

5359 5360 5361 5362 5363 5364
	/*
	 * If in_iowait is set, the code below may not trigger any cpufreq
	 * utilization updates, so do it here explicitly with the IOWAIT flag
	 * passed.
	 */
	if (p->in_iowait)
5365
		cpufreq_update_util(rq, SCHED_CPUFREQ_IOWAIT);
5366

5367
	for_each_sched_entity(se) {
5368
		if (se->on_rq)
5369 5370
			break;
		cfs_rq = cfs_rq_of(se);
5371
		enqueue_entity(cfs_rq, se, flags);
5372 5373 5374 5375 5376 5377

		/*
		 * end evaluation on encountering a throttled cfs_rq
		 *
		 * note: in the case of encountering a throttled cfs_rq we will
		 * post the final h_nr_running increment below.
5378
		 */
5379 5380
		if (cfs_rq_throttled(cfs_rq))
			break;
5381
		cfs_rq->h_nr_running++;
5382

5383
		flags = ENQUEUE_WAKEUP;
5384
	}
P
Peter Zijlstra 已提交
5385

P
Peter Zijlstra 已提交
5386
	for_each_sched_entity(se) {
5387
		cfs_rq = cfs_rq_of(se);
5388
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
5389

5390 5391 5392
		if (cfs_rq_throttled(cfs_rq))
			break;

5393
		update_load_avg(cfs_rq, se, UPDATE_TG);
5394
		update_cfs_group(se);
P
Peter Zijlstra 已提交
5395 5396
	}

Y
Yuyang Du 已提交
5397
	if (!se)
5398
		add_nr_running(rq, 1);
Y
Yuyang Du 已提交
5399

5400
	util_est_enqueue(&rq->cfs, p);
5401
	hrtick_update(rq);
5402 5403
}

5404 5405
static void set_next_buddy(struct sched_entity *se);

5406 5407 5408 5409 5410
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
5411
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
5412 5413
{
	struct cfs_rq *cfs_rq;
5414
	struct sched_entity *se = &p->se;
5415
	int task_sleep = flags & DEQUEUE_SLEEP;
5416 5417 5418

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
5419
		dequeue_entity(cfs_rq, se, flags);
5420 5421 5422 5423 5424 5425 5426 5427 5428

		/*
		 * end evaluation on encountering a throttled cfs_rq
		 *
		 * note: in the case of encountering a throttled cfs_rq we will
		 * post the final h_nr_running decrement below.
		*/
		if (cfs_rq_throttled(cfs_rq))
			break;
5429
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
5430

5431
		/* Don't dequeue parent if it has other entities besides us */
5432
		if (cfs_rq->load.weight) {
5433 5434
			/* Avoid re-evaluating load for this entity: */
			se = parent_entity(se);
5435 5436 5437 5438
			/*
			 * Bias pick_next to pick a task from this cfs_rq, as
			 * p is sleeping when it is within its sched_slice.
			 */
5439 5440
			if (task_sleep && se && !throttled_hierarchy(cfs_rq))
				set_next_buddy(se);
5441
			break;
5442
		}
5443
		flags |= DEQUEUE_SLEEP;
5444
	}
P
Peter Zijlstra 已提交
5445

P
Peter Zijlstra 已提交
5446
	for_each_sched_entity(se) {
5447
		cfs_rq = cfs_rq_of(se);
5448
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
5449

5450 5451 5452
		if (cfs_rq_throttled(cfs_rq))
			break;

5453
		update_load_avg(cfs_rq, se, UPDATE_TG);
5454
		update_cfs_group(se);
P
Peter Zijlstra 已提交
5455 5456
	}

Y
Yuyang Du 已提交
5457
	if (!se)
5458
		sub_nr_running(rq, 1);
Y
Yuyang Du 已提交
5459

5460
	util_est_dequeue(&rq->cfs, p, task_sleep);
5461
	hrtick_update(rq);
5462 5463
}

5464
#ifdef CONFIG_SMP
5465 5466 5467 5468 5469

/* Working cpumask for: load_balance, load_balance_newidle. */
DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
DEFINE_PER_CPU(cpumask_var_t, select_idle_mask);

5470
#ifdef CONFIG_NO_HZ_COMMON
5471 5472 5473 5474 5475
/*
 * per rq 'load' arrray crap; XXX kill this.
 */

/*
5476
 * The exact cpuload calculated at every tick would be:
5477
 *
5478 5479
 *   load' = (1 - 1/2^i) * load + (1/2^i) * cur_load
 *
5480 5481
 * If a CPU misses updates for n ticks (as it was idle) and update gets
 * called on the n+1-th tick when CPU may be busy, then we have:
5482 5483 5484
 *
 *   load_n   = (1 - 1/2^i)^n * load_0
 *   load_n+1 = (1 - 1/2^i)   * load_n + (1/2^i) * cur_load
5485 5486 5487
 *
 * decay_load_missed() below does efficient calculation of
 *
5488 5489 5490 5491 5492 5493
 *   load' = (1 - 1/2^i)^n * load
 *
 * Because x^(n+m) := x^n * x^m we can decompose any x^n in power-of-2 factors.
 * This allows us to precompute the above in said factors, thereby allowing the
 * reduction of an arbitrary n in O(log_2 n) steps. (See also
 * fixed_power_int())
5494
 *
5495
 * The calculation is approximated on a 128 point scale.
5496 5497
 */
#define DEGRADE_SHIFT		7
5498 5499 5500 5501 5502 5503 5504 5505 5506

static const u8 degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
static const u8 degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
	{   0,   0,  0,  0,  0,  0, 0, 0 },
	{  64,  32,  8,  0,  0,  0, 0, 0 },
	{  96,  72, 40, 12,  1,  0, 0, 0 },
	{ 112,  98, 75, 43, 15,  1, 0, 0 },
	{ 120, 112, 98, 76, 45, 16, 2, 0 }
};
5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535

/*
 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
 * would be when CPU is idle and so we just decay the old load without
 * adding any new load.
 */
static unsigned long
decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
{
	int j = 0;

	if (!missed_updates)
		return load;

	if (missed_updates >= degrade_zero_ticks[idx])
		return 0;

	if (idx == 1)
		return load >> missed_updates;

	while (missed_updates) {
		if (missed_updates % 2)
			load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;

		missed_updates >>= 1;
		j++;
	}
	return load;
}
5536 5537 5538 5539

static struct {
	cpumask_var_t idle_cpus_mask;
	atomic_t nr_cpus;
5540
	int has_blocked;		/* Idle CPUS has blocked load */
5541
	unsigned long next_balance;     /* in jiffy units */
5542
	unsigned long next_blocked;	/* Next update of blocked load in jiffies */
5543 5544
} nohz ____cacheline_aligned;

5545
#endif /* CONFIG_NO_HZ_COMMON */
5546

5547
/**
5548
 * __cpu_load_update - update the rq->cpu_load[] statistics
5549 5550 5551 5552
 * @this_rq: The rq to update statistics for
 * @this_load: The current load
 * @pending_updates: The number of missed updates
 *
5553
 * Update rq->cpu_load[] statistics. This function is usually called every
5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579
 * scheduler tick (TICK_NSEC).
 *
 * This function computes a decaying average:
 *
 *   load[i]' = (1 - 1/2^i) * load[i] + (1/2^i) * load
 *
 * Because of NOHZ it might not get called on every tick which gives need for
 * the @pending_updates argument.
 *
 *   load[i]_n = (1 - 1/2^i) * load[i]_n-1 + (1/2^i) * load_n-1
 *             = A * load[i]_n-1 + B ; A := (1 - 1/2^i), B := (1/2^i) * load
 *             = A * (A * load[i]_n-2 + B) + B
 *             = A * (A * (A * load[i]_n-3 + B) + B) + B
 *             = A^3 * load[i]_n-3 + (A^2 + A + 1) * B
 *             = A^n * load[i]_0 + (A^(n-1) + A^(n-2) + ... + 1) * B
 *             = A^n * load[i]_0 + ((1 - A^n) / (1 - A)) * B
 *             = (1 - 1/2^i)^n * (load[i]_0 - load) + load
 *
 * In the above we've assumed load_n := load, which is true for NOHZ_FULL as
 * any change in load would have resulted in the tick being turned back on.
 *
 * For regular NOHZ, this reduces to:
 *
 *   load[i]_n = (1 - 1/2^i)^n * load[i]_0
 *
 * see decay_load_misses(). For NOHZ_FULL we get to subtract and add the extra
5580
 * term.
5581
 */
5582 5583
static void cpu_load_update(struct rq *this_rq, unsigned long this_load,
			    unsigned long pending_updates)
5584
{
5585
	unsigned long __maybe_unused tickless_load = this_rq->cpu_load[0];
5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596
	int i, scale;

	this_rq->nr_load_updates++;

	/* Update our load: */
	this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
	for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
		unsigned long old_load, new_load;

		/* scale is effectively 1 << i now, and >> i divides by scale */

5597
		old_load = this_rq->cpu_load[i];
5598
#ifdef CONFIG_NO_HZ_COMMON
5599
		old_load = decay_load_missed(old_load, pending_updates - 1, i);
5600 5601 5602 5603 5604 5605 5606 5607 5608
		if (tickless_load) {
			old_load -= decay_load_missed(tickless_load, pending_updates - 1, i);
			/*
			 * old_load can never be a negative value because a
			 * decayed tickless_load cannot be greater than the
			 * original tickless_load.
			 */
			old_load += tickless_load;
		}
5609
#endif
5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624
		new_load = this_load;
		/*
		 * Round up the averaging division if load is increasing. This
		 * prevents us from getting stuck on 9 if the load is 10, for
		 * example.
		 */
		if (new_load > old_load)
			new_load += scale - 1;

		this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
	}

	sched_avg_update(this_rq);
}

5625
/* Used instead of source_load when we know the type == 0 */
5626
static unsigned long weighted_cpuload(struct rq *rq)
5627
{
5628
	return cfs_rq_runnable_load_avg(&rq->cfs);
5629 5630
}

5631
#ifdef CONFIG_NO_HZ_COMMON
5632 5633
/*
 * There is no sane way to deal with nohz on smp when using jiffies because the
5634
 * CPU doing the jiffies update might drift wrt the CPU doing the jiffy reading
5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648
 * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}.
 *
 * Therefore we need to avoid the delta approach from the regular tick when
 * possible since that would seriously skew the load calculation. This is why we
 * use cpu_load_update_periodic() for CPUs out of nohz. However we'll rely on
 * jiffies deltas for updates happening while in nohz mode (idle ticks, idle
 * loop exit, nohz_idle_balance, nohz full exit...)
 *
 * This means we might still be one tick off for nohz periods.
 */

static void cpu_load_update_nohz(struct rq *this_rq,
				 unsigned long curr_jiffies,
				 unsigned long load)
5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659
{
	unsigned long pending_updates;

	pending_updates = curr_jiffies - this_rq->last_load_update_tick;
	if (pending_updates) {
		this_rq->last_load_update_tick = curr_jiffies;
		/*
		 * In the regular NOHZ case, we were idle, this means load 0.
		 * In the NOHZ_FULL case, we were non-idle, we should consider
		 * its weighted load.
		 */
5660
		cpu_load_update(this_rq, load, pending_updates);
5661 5662 5663
	}
}

5664 5665 5666 5667
/*
 * Called from nohz_idle_balance() to update the load ratings before doing the
 * idle balance.
 */
5668
static void cpu_load_update_idle(struct rq *this_rq)
5669 5670 5671 5672
{
	/*
	 * bail if there's load or we're actually up-to-date.
	 */
5673
	if (weighted_cpuload(this_rq))
5674 5675
		return;

5676
	cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), 0);
5677 5678 5679
}

/*
5680 5681 5682 5683
 * Record CPU load on nohz entry so we know the tickless load to account
 * on nohz exit. cpu_load[0] happens then to be updated more frequently
 * than other cpu_load[idx] but it should be fine as cpu_load readers
 * shouldn't rely into synchronized cpu_load[*] updates.
5684
 */
5685
void cpu_load_update_nohz_start(void)
5686 5687
{
	struct rq *this_rq = this_rq();
5688 5689 5690 5691 5692 5693

	/*
	 * This is all lockless but should be fine. If weighted_cpuload changes
	 * concurrently we'll exit nohz. And cpu_load write can race with
	 * cpu_load_update_idle() but both updater would be writing the same.
	 */
5694
	this_rq->cpu_load[0] = weighted_cpuload(this_rq);
5695 5696 5697 5698 5699 5700 5701
}

/*
 * Account the tickless load in the end of a nohz frame.
 */
void cpu_load_update_nohz_stop(void)
{
5702
	unsigned long curr_jiffies = READ_ONCE(jiffies);
5703 5704
	struct rq *this_rq = this_rq();
	unsigned long load;
5705
	struct rq_flags rf;
5706 5707 5708 5709

	if (curr_jiffies == this_rq->last_load_update_tick)
		return;

5710
	load = weighted_cpuload(this_rq);
5711
	rq_lock(this_rq, &rf);
5712
	update_rq_clock(this_rq);
5713
	cpu_load_update_nohz(this_rq, curr_jiffies, load);
5714
	rq_unlock(this_rq, &rf);
5715
}
5716 5717 5718 5719 5720 5721 5722 5723
#else /* !CONFIG_NO_HZ_COMMON */
static inline void cpu_load_update_nohz(struct rq *this_rq,
					unsigned long curr_jiffies,
					unsigned long load) { }
#endif /* CONFIG_NO_HZ_COMMON */

static void cpu_load_update_periodic(struct rq *this_rq, unsigned long load)
{
5724
#ifdef CONFIG_NO_HZ_COMMON
5725 5726
	/* See the mess around cpu_load_update_nohz(). */
	this_rq->last_load_update_tick = READ_ONCE(jiffies);
5727
#endif
5728 5729
	cpu_load_update(this_rq, load, 1);
}
5730 5731 5732 5733

/*
 * Called from scheduler_tick()
 */
5734
void cpu_load_update_active(struct rq *this_rq)
5735
{
5736
	unsigned long load = weighted_cpuload(this_rq);
5737 5738 5739 5740 5741

	if (tick_nohz_tick_stopped())
		cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), load);
	else
		cpu_load_update_periodic(this_rq, load);
5742 5743
}

5744
/*
5745
 * Return a low guess at the load of a migration-source CPU weighted
5746 5747 5748 5749 5750 5751 5752 5753
 * according to the scheduling class and "nice" value.
 *
 * We want to under-estimate the load of migration sources, to
 * balance conservatively.
 */
static unsigned long source_load(int cpu, int type)
{
	struct rq *rq = cpu_rq(cpu);
5754
	unsigned long total = weighted_cpuload(rq);
5755 5756 5757 5758 5759 5760 5761 5762

	if (type == 0 || !sched_feat(LB_BIAS))
		return total;

	return min(rq->cpu_load[type-1], total);
}

/*
5763
 * Return a high guess at the load of a migration-target CPU weighted
5764 5765 5766 5767 5768
 * according to the scheduling class and "nice" value.
 */
static unsigned long target_load(int cpu, int type)
{
	struct rq *rq = cpu_rq(cpu);
5769
	unsigned long total = weighted_cpuload(rq);
5770 5771 5772 5773 5774 5775 5776

	if (type == 0 || !sched_feat(LB_BIAS))
		return total;

	return max(rq->cpu_load[type-1], total);
}

5777
static unsigned long capacity_of(int cpu)
5778
{
5779
	return cpu_rq(cpu)->cpu_capacity;
5780 5781
}

5782 5783 5784 5785 5786
static unsigned long capacity_orig_of(int cpu)
{
	return cpu_rq(cpu)->cpu_capacity_orig;
}

5787 5788 5789
static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
5790
	unsigned long nr_running = READ_ONCE(rq->cfs.h_nr_running);
5791
	unsigned long load_avg = weighted_cpuload(rq);
5792 5793

	if (nr_running)
5794
		return load_avg / nr_running;
5795 5796 5797 5798

	return 0;
}

P
Peter Zijlstra 已提交
5799 5800 5801 5802 5803 5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815
static void record_wakee(struct task_struct *p)
{
	/*
	 * Only decay a single time; tasks that have less then 1 wakeup per
	 * jiffy will not have built up many flips.
	 */
	if (time_after(jiffies, current->wakee_flip_decay_ts + HZ)) {
		current->wakee_flips >>= 1;
		current->wakee_flip_decay_ts = jiffies;
	}

	if (current->last_wakee != p) {
		current->last_wakee = p;
		current->wakee_flips++;
	}
}

M
Mike Galbraith 已提交
5816 5817
/*
 * Detect M:N waker/wakee relationships via a switching-frequency heuristic.
P
Peter Zijlstra 已提交
5818
 *
M
Mike Galbraith 已提交
5819
 * A waker of many should wake a different task than the one last awakened
P
Peter Zijlstra 已提交
5820 5821 5822 5823 5824 5825 5826 5827 5828 5829 5830 5831
 * at a frequency roughly N times higher than one of its wakees.
 *
 * In order to determine whether we should let the load spread vs consolidating
 * to shared cache, we look for a minimum 'flip' frequency of llc_size in one
 * partner, and a factor of lls_size higher frequency in the other.
 *
 * With both conditions met, we can be relatively sure that the relationship is
 * non-monogamous, with partner count exceeding socket size.
 *
 * Waker/wakee being client/server, worker/dispatcher, interrupt source or
 * whatever is irrelevant, spread criteria is apparent partner count exceeds
 * socket size.
M
Mike Galbraith 已提交
5832
 */
5833 5834
static int wake_wide(struct task_struct *p)
{
M
Mike Galbraith 已提交
5835 5836
	unsigned int master = current->wakee_flips;
	unsigned int slave = p->wakee_flips;
5837
	int factor = this_cpu_read(sd_llc_size);
5838

M
Mike Galbraith 已提交
5839 5840 5841 5842 5843
	if (master < slave)
		swap(master, slave);
	if (slave < factor || master < slave * factor)
		return 0;
	return 1;
5844 5845
}

5846
/*
5847 5848 5849
 * The purpose of wake_affine() is to quickly determine on which CPU we can run
 * soonest. For the purpose of speed we only consider the waking and previous
 * CPU.
5850
 *
5851 5852
 * wake_affine_idle() - only considers 'now', it check if the waking CPU is
 *			cache-affine and is (or	will be) idle.
5853 5854 5855 5856
 *
 * wake_affine_weight() - considers the weight to reflect the average
 *			  scheduling latency of the CPUs. This seems to work
 *			  for the overloaded case.
5857
 */
5858
static int
5859
wake_affine_idle(int this_cpu, int prev_cpu, int sync)
5860
{
5861 5862 5863 5864 5865
	/*
	 * If this_cpu is idle, it implies the wakeup is from interrupt
	 * context. Only allow the move if cache is shared. Otherwise an
	 * interrupt intensive workload could force all tasks onto one
	 * node depending on the IO topology or IRQ affinity settings.
5866 5867 5868 5869 5870 5871
	 *
	 * If the prev_cpu is idle and cache affine then avoid a migration.
	 * There is no guarantee that the cache hot data from an interrupt
	 * is more important than cache hot data on the prev_cpu and from
	 * a cpufreq perspective, it's better to have higher utilisation
	 * on one CPU.
5872 5873
	 */
	if (idle_cpu(this_cpu) && cpus_share_cache(this_cpu, prev_cpu))
5874
		return idle_cpu(prev_cpu) ? prev_cpu : this_cpu;
5875

5876
	if (sync && cpu_rq(this_cpu)->nr_running == 1)
5877
		return this_cpu;
5878

5879
	return nr_cpumask_bits;
5880 5881
}

5882
static int
5883 5884
wake_affine_weight(struct sched_domain *sd, struct task_struct *p,
		   int this_cpu, int prev_cpu, int sync)
5885 5886 5887 5888
{
	s64 this_eff_load, prev_eff_load;
	unsigned long task_load;

5889
	this_eff_load = target_load(this_cpu, sd->wake_idx);
5890 5891 5892 5893

	if (sync) {
		unsigned long current_load = task_h_load(current);

5894
		if (current_load > this_eff_load)
5895
			return this_cpu;
5896

5897
		this_eff_load -= current_load;
5898 5899 5900 5901
	}

	task_load = task_h_load(p);

5902 5903 5904 5905
	this_eff_load += task_load;
	if (sched_feat(WA_BIAS))
		this_eff_load *= 100;
	this_eff_load *= capacity_of(prev_cpu);
5906

5907
	prev_eff_load = source_load(prev_cpu, sd->wake_idx);
5908 5909 5910 5911
	prev_eff_load -= task_load;
	if (sched_feat(WA_BIAS))
		prev_eff_load *= 100 + (sd->imbalance_pct - 100) / 2;
	prev_eff_load *= capacity_of(this_cpu);
5912

5913 5914 5915 5916 5917 5918 5919 5920 5921 5922
	/*
	 * If sync, adjust the weight of prev_eff_load such that if
	 * prev_eff == this_eff that select_idle_sibling() will consider
	 * stacking the wakee on top of the waker if no other CPU is
	 * idle.
	 */
	if (sync)
		prev_eff_load += 1;

	return this_eff_load < prev_eff_load ? this_cpu : nr_cpumask_bits;
5923 5924
}

5925 5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966
#ifdef CONFIG_NUMA_BALANCING
static void
update_wa_numa_placement(struct task_struct *p, int prev_cpu, int target)
{
	unsigned long interval;

	if (!static_branch_likely(&sched_numa_balancing))
		return;

	/* If balancing has no preference then continue gathering data */
	if (p->numa_preferred_nid == -1)
		return;

	/*
	 * If the wakeup is not affecting locality then it is neutral from
	 * the perspective of NUMA balacing so continue gathering data.
	 */
	if (cpu_to_node(prev_cpu) == cpu_to_node(target))
		return;

	/*
	 * Temporarily prevent NUMA balancing trying to place waker/wakee after
	 * wakee has been moved by wake_affine. This will potentially allow
	 * related tasks to converge and update their data placement. The
	 * 4 * numa_scan_period is to allow the two-pass filter to migrate
	 * hot data to the wakers node.
	 */
	interval = max(sysctl_numa_balancing_scan_delay,
			 p->numa_scan_period << 2);
	p->numa_migrate_retry = jiffies + msecs_to_jiffies(interval);

	interval = max(sysctl_numa_balancing_scan_delay,
			 current->numa_scan_period << 2);
	current->numa_migrate_retry = jiffies + msecs_to_jiffies(interval);
}
#else
static void
update_wa_numa_placement(struct task_struct *p, int prev_cpu, int target)
{
}
#endif

5967
static int wake_affine(struct sched_domain *sd, struct task_struct *p,
5968
		       int this_cpu, int prev_cpu, int sync)
5969
{
5970
	int target = nr_cpumask_bits;
5971

5972
	if (sched_feat(WA_IDLE))
5973
		target = wake_affine_idle(this_cpu, prev_cpu, sync);
5974

5975 5976
	if (sched_feat(WA_WEIGHT) && target == nr_cpumask_bits)
		target = wake_affine_weight(sd, p, this_cpu, prev_cpu, sync);
5977

5978
	schedstat_inc(p->se.statistics.nr_wakeups_affine_attempts);
5979 5980
	if (target == nr_cpumask_bits)
		return prev_cpu;
5981

5982
	update_wa_numa_placement(p, prev_cpu, target);
5983 5984 5985
	schedstat_inc(sd->ttwu_move_affine);
	schedstat_inc(p->se.statistics.nr_wakeups_affine);
	return target;
5986 5987
}

5988
static unsigned long cpu_util_wake(int cpu, struct task_struct *p);
5989 5990 5991

static unsigned long capacity_spare_wake(int cpu, struct task_struct *p)
{
5992
	return max_t(long, capacity_of(cpu) - cpu_util_wake(cpu, p), 0);
5993 5994
}

5995 5996 5997
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
5998 5999
 *
 * Assumes p is allowed on at least one CPU in sd.
6000 6001
 */
static struct sched_group *
P
Peter Zijlstra 已提交
6002
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
6003
		  int this_cpu, int sd_flag)
6004
{
6005
	struct sched_group *idlest = NULL, *group = sd->groups;
6006
	struct sched_group *most_spare_sg = NULL;
6007 6008 6009
	unsigned long min_runnable_load = ULONG_MAX;
	unsigned long this_runnable_load = ULONG_MAX;
	unsigned long min_avg_load = ULONG_MAX, this_avg_load = ULONG_MAX;
6010
	unsigned long most_spare = 0, this_spare = 0;
6011
	int load_idx = sd->forkexec_idx;
6012 6013 6014
	int imbalance_scale = 100 + (sd->imbalance_pct-100)/2;
	unsigned long imbalance = scale_load_down(NICE_0_LOAD) *
				(sd->imbalance_pct-100) / 100;
6015

6016 6017 6018
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

6019
	do {
6020 6021
		unsigned long load, avg_load, runnable_load;
		unsigned long spare_cap, max_spare_cap;
6022 6023
		int local_group;
		int i;
6024

6025
		/* Skip over this group if it has no CPUs allowed */
6026
		if (!cpumask_intersects(sched_group_span(group),
6027
					&p->cpus_allowed))
6028 6029 6030
			continue;

		local_group = cpumask_test_cpu(this_cpu,
6031
					       sched_group_span(group));
6032

6033 6034 6035 6036
		/*
		 * Tally up the load of all CPUs in the group and find
		 * the group containing the CPU with most spare capacity.
		 */
6037
		avg_load = 0;
6038
		runnable_load = 0;
6039
		max_spare_cap = 0;
6040

6041
		for_each_cpu(i, sched_group_span(group)) {
6042
			/* Bias balancing toward CPUs of our domain */
6043 6044 6045 6046 6047
			if (local_group)
				load = source_load(i, load_idx);
			else
				load = target_load(i, load_idx);

6048 6049 6050
			runnable_load += load;

			avg_load += cfs_rq_load_avg(&cpu_rq(i)->cfs);
6051 6052 6053 6054 6055

			spare_cap = capacity_spare_wake(i, p);

			if (spare_cap > max_spare_cap)
				max_spare_cap = spare_cap;
6056 6057
		}

6058
		/* Adjust by relative CPU capacity of the group */
6059 6060 6061 6062
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) /
					group->sgc->capacity;
		runnable_load = (runnable_load * SCHED_CAPACITY_SCALE) /
					group->sgc->capacity;
6063 6064

		if (local_group) {
6065 6066
			this_runnable_load = runnable_load;
			this_avg_load = avg_load;
6067 6068
			this_spare = max_spare_cap;
		} else {
6069 6070 6071
			if (min_runnable_load > (runnable_load + imbalance)) {
				/*
				 * The runnable load is significantly smaller
6072
				 * so we can pick this new CPU:
6073 6074 6075 6076 6077 6078 6079 6080
				 */
				min_runnable_load = runnable_load;
				min_avg_load = avg_load;
				idlest = group;
			} else if ((runnable_load < (min_runnable_load + imbalance)) &&
				   (100*min_avg_load > imbalance_scale*avg_load)) {
				/*
				 * The runnable loads are close so take the
6081
				 * blocked load into account through avg_load:
6082 6083
				 */
				min_avg_load = avg_load;
6084 6085 6086 6087 6088 6089 6090
				idlest = group;
			}

			if (most_spare < max_spare_cap) {
				most_spare = max_spare_cap;
				most_spare_sg = group;
			}
6091 6092 6093
		}
	} while (group = group->next, group != sd->groups);

6094 6095 6096 6097 6098 6099
	/*
	 * The cross-over point between using spare capacity or least load
	 * is too conservative for high utilization tasks on partially
	 * utilized systems if we require spare_capacity > task_util(p),
	 * so we allow for some task stuffing by using
	 * spare_capacity > task_util(p)/2.
6100 6101 6102 6103
	 *
	 * Spare capacity can't be used for fork because the utilization has
	 * not been set yet, we must first select a rq to compute the initial
	 * utilization.
6104
	 */
6105 6106 6107
	if (sd_flag & SD_BALANCE_FORK)
		goto skip_spare;

6108
	if (this_spare > task_util(p) / 2 &&
6109
	    imbalance_scale*this_spare > 100*most_spare)
6110
		return NULL;
6111 6112

	if (most_spare > task_util(p) / 2)
6113 6114
		return most_spare_sg;

6115
skip_spare:
6116 6117 6118
	if (!idlest)
		return NULL;

6119 6120 6121 6122 6123 6124 6125 6126 6127 6128 6129 6130
	/*
	 * When comparing groups across NUMA domains, it's possible for the
	 * local domain to be very lightly loaded relative to the remote
	 * domains but "imbalance" skews the comparison making remote CPUs
	 * look much more favourable. When considering cross-domain, add
	 * imbalance to the runnable load on the remote node and consider
	 * staying local.
	 */
	if ((sd->flags & SD_NUMA) &&
	    min_runnable_load + imbalance >= this_runnable_load)
		return NULL;

6131
	if (min_runnable_load > (this_runnable_load + imbalance))
6132
		return NULL;
6133 6134 6135 6136 6137

	if ((this_runnable_load < (min_runnable_load + imbalance)) &&
	     (100*this_avg_load < imbalance_scale*min_avg_load))
		return NULL;

6138 6139 6140 6141
	return idlest;
}

/*
6142
 * find_idlest_group_cpu - find the idlest CPU among the CPUs in the group.
6143 6144
 */
static int
6145
find_idlest_group_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
6146 6147
{
	unsigned long load, min_load = ULONG_MAX;
6148 6149 6150 6151
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
6152 6153
	int i;

6154 6155
	/* Check if we have any choice: */
	if (group->group_weight == 1)
6156
		return cpumask_first(sched_group_span(group));
6157

6158
	/* Traverse only the allowed CPUs */
6159
	for_each_cpu_and(i, sched_group_span(group), &p->cpus_allowed) {
6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175 6176 6177 6178 6179 6180 6181
		if (idle_cpu(i)) {
			struct rq *rq = cpu_rq(i);
			struct cpuidle_state *idle = idle_get_state(rq);
			if (idle && idle->exit_latency < min_exit_latency) {
				/*
				 * We give priority to a CPU whose idle state
				 * has the smallest exit latency irrespective
				 * of any idle timestamp.
				 */
				min_exit_latency = idle->exit_latency;
				latest_idle_timestamp = rq->idle_stamp;
				shallowest_idle_cpu = i;
			} else if ((!idle || idle->exit_latency == min_exit_latency) &&
				   rq->idle_stamp > latest_idle_timestamp) {
				/*
				 * If equal or no active idle state, then
				 * the most recently idled CPU might have
				 * a warmer cache.
				 */
				latest_idle_timestamp = rq->idle_stamp;
				shallowest_idle_cpu = i;
			}
6182
		} else if (shallowest_idle_cpu == -1) {
6183
			load = weighted_cpuload(cpu_rq(i));
6184
			if (load < min_load) {
6185 6186 6187
				min_load = load;
				least_loaded_cpu = i;
			}
6188 6189 6190
		}
	}

6191
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
6192
}
6193

6194 6195 6196
static inline int find_idlest_cpu(struct sched_domain *sd, struct task_struct *p,
				  int cpu, int prev_cpu, int sd_flag)
{
6197
	int new_cpu = cpu;
6198

6199 6200 6201
	if (!cpumask_intersects(sched_domain_span(sd), &p->cpus_allowed))
		return prev_cpu;

6202 6203 6204 6205 6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216 6217 6218
	while (sd) {
		struct sched_group *group;
		struct sched_domain *tmp;
		int weight;

		if (!(sd->flags & sd_flag)) {
			sd = sd->child;
			continue;
		}

		group = find_idlest_group(sd, p, cpu, sd_flag);
		if (!group) {
			sd = sd->child;
			continue;
		}

		new_cpu = find_idlest_group_cpu(group, p, cpu);
6219
		if (new_cpu == cpu) {
6220
			/* Now try balancing at a lower domain level of 'cpu': */
6221 6222 6223 6224
			sd = sd->child;
			continue;
		}

6225
		/* Now try balancing at a lower domain level of 'new_cpu': */
6226 6227 6228 6229 6230 6231 6232 6233 6234 6235 6236 6237 6238 6239
		cpu = new_cpu;
		weight = sd->span_weight;
		sd = NULL;
		for_each_domain(cpu, tmp) {
			if (weight <= tmp->span_weight)
				break;
			if (tmp->flags & sd_flag)
				sd = tmp;
		}
	}

	return new_cpu;
}

6240 6241 6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256 6257 6258 6259 6260 6261 6262 6263 6264 6265 6266 6267 6268
#ifdef CONFIG_SCHED_SMT

static inline void set_idle_cores(int cpu, int val)
{
	struct sched_domain_shared *sds;

	sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));
	if (sds)
		WRITE_ONCE(sds->has_idle_cores, val);
}

static inline bool test_idle_cores(int cpu, bool def)
{
	struct sched_domain_shared *sds;

	sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));
	if (sds)
		return READ_ONCE(sds->has_idle_cores);

	return def;
}

/*
 * Scans the local SMT mask to see if the entire core is idle, and records this
 * information in sd_llc_shared->has_idle_cores.
 *
 * Since SMT siblings share all cache levels, inspecting this limited remote
 * state should be fairly cheap.
 */
P
Peter Zijlstra 已提交
6269
void __update_idle_core(struct rq *rq)
6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296 6297 6298
{
	int core = cpu_of(rq);
	int cpu;

	rcu_read_lock();
	if (test_idle_cores(core, true))
		goto unlock;

	for_each_cpu(cpu, cpu_smt_mask(core)) {
		if (cpu == core)
			continue;

		if (!idle_cpu(cpu))
			goto unlock;
	}

	set_idle_cores(core, 1);
unlock:
	rcu_read_unlock();
}

/*
 * Scan the entire LLC domain for idle cores; this dynamically switches off if
 * there are no idle cores left in the system; tracked through
 * sd_llc->shared->has_idle_cores and enabled through update_idle_core() above.
 */
static int select_idle_core(struct task_struct *p, struct sched_domain *sd, int target)
{
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(select_idle_mask);
6299
	int core, cpu;
6300

P
Peter Zijlstra 已提交
6301 6302 6303
	if (!static_branch_likely(&sched_smt_present))
		return -1;

6304 6305 6306
	if (!test_idle_cores(target, false))
		return -1;

6307
	cpumask_and(cpus, sched_domain_span(sd), &p->cpus_allowed);
6308

6309
	for_each_cpu_wrap(core, cpus, target) {
6310 6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333 6334 6335 6336
		bool idle = true;

		for_each_cpu(cpu, cpu_smt_mask(core)) {
			cpumask_clear_cpu(cpu, cpus);
			if (!idle_cpu(cpu))
				idle = false;
		}

		if (idle)
			return core;
	}

	/*
	 * Failed to find an idle core; stop looking for one.
	 */
	set_idle_cores(target, 0);

	return -1;
}

/*
 * Scan the local SMT mask for idle CPUs.
 */
static int select_idle_smt(struct task_struct *p, struct sched_domain *sd, int target)
{
	int cpu;

P
Peter Zijlstra 已提交
6337 6338 6339
	if (!static_branch_likely(&sched_smt_present))
		return -1;

6340
	for_each_cpu(cpu, cpu_smt_mask(target)) {
6341
		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
6342 6343 6344 6345 6346 6347 6348 6349 6350 6351 6352 6353 6354 6355 6356 6357 6358 6359 6360 6361 6362 6363 6364 6365 6366 6367
			continue;
		if (idle_cpu(cpu))
			return cpu;
	}

	return -1;
}

#else /* CONFIG_SCHED_SMT */

static inline int select_idle_core(struct task_struct *p, struct sched_domain *sd, int target)
{
	return -1;
}

static inline int select_idle_smt(struct task_struct *p, struct sched_domain *sd, int target)
{
	return -1;
}

#endif /* CONFIG_SCHED_SMT */

/*
 * Scan the LLC domain for idle CPUs; this is dynamically regulated by
 * comparing the average scan cost (tracked in sd->avg_scan_cost) against the
 * average idle time for this rq (as found in rq->avg_idle).
6368
 */
6369 6370
static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int target)
{
6371
	struct sched_domain *this_sd;
6372
	u64 avg_cost, avg_idle;
6373 6374
	u64 time, cost;
	s64 delta;
6375
	int cpu, nr = INT_MAX;
6376

6377 6378 6379 6380
	this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
	if (!this_sd)
		return -1;

6381 6382 6383 6384
	/*
	 * Due to large variance we need a large fuzz factor; hackbench in
	 * particularly is sensitive here.
	 */
6385 6386 6387 6388
	avg_idle = this_rq()->avg_idle / 512;
	avg_cost = this_sd->avg_scan_cost + 1;

	if (sched_feat(SIS_AVG_CPU) && avg_idle < avg_cost)
6389 6390
		return -1;

6391 6392 6393 6394 6395 6396 6397 6398
	if (sched_feat(SIS_PROP)) {
		u64 span_avg = sd->span_weight * avg_idle;
		if (span_avg > 4*avg_cost)
			nr = div_u64(span_avg, avg_cost);
		else
			nr = 4;
	}

6399 6400
	time = local_clock();

6401
	for_each_cpu_wrap(cpu, sched_domain_span(sd), target) {
6402 6403
		if (!--nr)
			return -1;
6404
		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
6405 6406 6407 6408 6409 6410 6411 6412 6413 6414 6415 6416 6417 6418 6419
			continue;
		if (idle_cpu(cpu))
			break;
	}

	time = local_clock() - time;
	cost = this_sd->avg_scan_cost;
	delta = (s64)(time - cost) / 8;
	this_sd->avg_scan_cost += delta;

	return cpu;
}

/*
 * Try and locate an idle core/thread in the LLC cache domain.
6420
 */
6421
static int select_idle_sibling(struct task_struct *p, int prev, int target)
6422
{
6423
	struct sched_domain *sd;
6424
	int i, recent_used_cpu;
6425

6426 6427
	if (idle_cpu(target))
		return target;
6428 6429

	/*
6430
	 * If the previous CPU is cache affine and idle, don't be stupid:
6431
	 */
6432 6433
	if (prev != target && cpus_share_cache(prev, target) && idle_cpu(prev))
		return prev;
6434

6435
	/* Check a recently used CPU as a potential idle candidate: */
6436 6437 6438 6439 6440 6441 6442 6443
	recent_used_cpu = p->recent_used_cpu;
	if (recent_used_cpu != prev &&
	    recent_used_cpu != target &&
	    cpus_share_cache(recent_used_cpu, target) &&
	    idle_cpu(recent_used_cpu) &&
	    cpumask_test_cpu(p->recent_used_cpu, &p->cpus_allowed)) {
		/*
		 * Replace recent_used_cpu with prev as it is a potential
6444
		 * candidate for the next wake:
6445 6446 6447 6448 6449
		 */
		p->recent_used_cpu = prev;
		return recent_used_cpu;
	}

6450
	sd = rcu_dereference(per_cpu(sd_llc, target));
6451 6452
	if (!sd)
		return target;
6453

6454 6455 6456
	i = select_idle_core(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;
6457

6458 6459 6460 6461 6462 6463 6464
	i = select_idle_cpu(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;

	i = select_idle_smt(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;
6465

6466 6467
	return target;
}
6468

6469 6470 6471 6472 6473 6474 6475
/**
 * Amount of capacity of a CPU that is (estimated to be) used by CFS tasks
 * @cpu: the CPU to get the utilization of
 *
 * The unit of the return value must be the one of capacity so we can compare
 * the utilization with the capacity of the CPU that is available for CFS task
 * (ie cpu_capacity).
6476 6477 6478 6479 6480 6481 6482 6483 6484 6485
 *
 * cfs_rq.avg.util_avg is the sum of running time of runnable tasks plus the
 * recent utilization of currently non-runnable tasks on a CPU. It represents
 * the amount of utilization of a CPU in the range [0..capacity_orig] where
 * capacity_orig is the cpu_capacity available at the highest frequency
 * (arch_scale_freq_capacity()).
 * The utilization of a CPU converges towards a sum equal to or less than the
 * current capacity (capacity_curr <= capacity_orig) of the CPU because it is
 * the running time on this CPU scaled by capacity_curr.
 *
6486 6487 6488 6489 6490 6491 6492 6493
 * The estimated utilization of a CPU is defined to be the maximum between its
 * cfs_rq.avg.util_avg and the sum of the estimated utilization of the tasks
 * currently RUNNABLE on that CPU.
 * This allows to properly represent the expected utilization of a CPU which
 * has just got a big task running since a long sleep period. At the same time
 * however it preserves the benefits of the "blocked utilization" in
 * describing the potential for other tasks waking up on the same CPU.
 *
6494 6495 6496 6497 6498 6499 6500 6501 6502 6503
 * Nevertheless, cfs_rq.avg.util_avg can be higher than capacity_curr or even
 * higher than capacity_orig because of unfortunate rounding in
 * cfs.avg.util_avg or just after migrating tasks and new task wakeups until
 * the average stabilizes with the new running time. We need to check that the
 * utilization stays within the range of [0..capacity_orig] and cap it if
 * necessary. Without utilization capping, a group could be seen as overloaded
 * (CPU0 utilization at 121% + CPU1 utilization at 80%) whereas CPU1 has 20% of
 * available capacity. We allow utilization to overshoot capacity_curr (but not
 * capacity_orig) as it useful for predicting the capacity required after task
 * migrations (scheduler-driven DVFS).
6504 6505
 *
 * Return: the (estimated) utilization for the specified CPU
6506
 */
6507
static inline unsigned long cpu_util(int cpu)
6508
{
6509 6510 6511 6512 6513 6514 6515 6516
	struct cfs_rq *cfs_rq;
	unsigned int util;

	cfs_rq = &cpu_rq(cpu)->cfs;
	util = READ_ONCE(cfs_rq->avg.util_avg);

	if (sched_feat(UTIL_EST))
		util = max(util, READ_ONCE(cfs_rq->avg.util_est.enqueued));
6517

6518
	return min_t(unsigned long, util, capacity_orig_of(cpu));
6519
}
6520

6521
/*
6522
 * cpu_util_wake: Compute CPU utilization with any contributions from
6523 6524
 * the waking task p removed.
 */
6525
static unsigned long cpu_util_wake(int cpu, struct task_struct *p)
6526
{
6527 6528
	struct cfs_rq *cfs_rq;
	unsigned int util;
6529 6530

	/* Task has no contribution or is new */
6531
	if (cpu != task_cpu(p) || !READ_ONCE(p->se.avg.last_update_time))
6532 6533
		return cpu_util(cpu);

6534 6535 6536 6537 6538
	cfs_rq = &cpu_rq(cpu)->cfs;
	util = READ_ONCE(cfs_rq->avg.util_avg);

	/* Discount task's blocked util from CPU's util */
	util -= min_t(unsigned int, util, task_util(p));
6539

6540 6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574
	/*
	 * Covered cases:
	 *
	 * a) if *p is the only task sleeping on this CPU, then:
	 *      cpu_util (== task_util) > util_est (== 0)
	 *    and thus we return:
	 *      cpu_util_wake = (cpu_util - task_util) = 0
	 *
	 * b) if other tasks are SLEEPING on this CPU, which is now exiting
	 *    IDLE, then:
	 *      cpu_util >= task_util
	 *      cpu_util > util_est (== 0)
	 *    and thus we discount *p's blocked utilization to return:
	 *      cpu_util_wake = (cpu_util - task_util) >= 0
	 *
	 * c) if other tasks are RUNNABLE on that CPU and
	 *      util_est > cpu_util
	 *    then we use util_est since it returns a more restrictive
	 *    estimation of the spare capacity on that CPU, by just
	 *    considering the expected utilization of tasks already
	 *    runnable on that CPU.
	 *
	 * Cases a) and b) are covered by the above code, while case c) is
	 * covered by the following code when estimated utilization is
	 * enabled.
	 */
	if (sched_feat(UTIL_EST))
		util = max(util, READ_ONCE(cfs_rq->avg.util_est.enqueued));

	/*
	 * Utilization (estimated) can exceed the CPU capacity, thus let's
	 * clamp to the maximum CPU capacity to ensure consistency with
	 * the cpu_util call.
	 */
	return min_t(unsigned long, util, capacity_orig_of(cpu));
6575 6576
}

6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587 6588 6589 6590 6591 6592 6593 6594
/*
 * Disable WAKE_AFFINE in the case where task @p doesn't fit in the
 * capacity of either the waking CPU @cpu or the previous CPU @prev_cpu.
 *
 * In that case WAKE_AFFINE doesn't make sense and we'll let
 * BALANCE_WAKE sort things out.
 */
static int wake_cap(struct task_struct *p, int cpu, int prev_cpu)
{
	long min_cap, max_cap;

	min_cap = min(capacity_orig_of(prev_cpu), capacity_orig_of(cpu));
	max_cap = cpu_rq(cpu)->rd->max_cpu_capacity;

	/* Minimum capacity is close to max, no need to abort wake_affine */
	if (max_cap - min_cap < max_cap >> 3)
		return 0;

6595 6596 6597
	/* Bring task utilization in sync with prev_cpu */
	sync_entity_load_avg(&p->se);

6598 6599 6600
	return min_cap * 1024 < task_util(p) * capacity_margin;
}

6601
/*
6602 6603 6604
 * select_task_rq_fair: Select target runqueue for the waking task in domains
 * that have the 'sd_flag' flag set. In practice, this is SD_BALANCE_WAKE,
 * SD_BALANCE_FORK, or SD_BALANCE_EXEC.
6605
 *
6606 6607
 * Balances load by selecting the idlest CPU in the idlest group, or under
 * certain conditions an idle sibling CPU if the domain has SD_WAKE_AFFINE set.
6608
 *
6609
 * Returns the target CPU number.
6610 6611 6612
 *
 * preempt must be disabled.
 */
6613
static int
6614
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
6615
{
6616
	struct sched_domain *tmp, *sd = NULL;
6617
	int cpu = smp_processor_id();
M
Mike Galbraith 已提交
6618
	int new_cpu = prev_cpu;
6619
	int want_affine = 0;
6620
	int sync = (wake_flags & WF_SYNC) && !(current->flags & PF_EXITING);
6621

P
Peter Zijlstra 已提交
6622 6623
	if (sd_flag & SD_BALANCE_WAKE) {
		record_wakee(p);
6624
		want_affine = !wake_wide(p) && !wake_cap(p, cpu, prev_cpu)
6625
			      && cpumask_test_cpu(cpu, &p->cpus_allowed);
P
Peter Zijlstra 已提交
6626
	}
6627

6628
	rcu_read_lock();
6629
	for_each_domain(cpu, tmp) {
6630
		if (!(tmp->flags & SD_LOAD_BALANCE))
M
Mike Galbraith 已提交
6631
			break;
6632

6633
		/*
6634
		 * If both 'cpu' and 'prev_cpu' are part of this domain,
6635
		 * cpu is a valid SD_WAKE_AFFINE target.
6636
		 */
6637 6638
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
6639 6640 6641 6642
			if (cpu != prev_cpu)
				new_cpu = wake_affine(tmp, p, cpu, prev_cpu, sync);

			sd = NULL; /* Prefer wake_affine over balance flags */
6643
			break;
6644
		}
6645

6646
		if (tmp->flags & sd_flag)
6647
			sd = tmp;
M
Mike Galbraith 已提交
6648 6649
		else if (!want_affine)
			break;
6650 6651
	}

6652 6653
	if (unlikely(sd)) {
		/* Slow path */
6654

6655 6656 6657 6658 6659
		/*
		 * We're going to need the task's util for capacity_spare_wake
		 * in find_idlest_group. Sync it up to prev_cpu's
		 * last_update_time.
		 */
6660 6661
		if (!(sd_flag & SD_BALANCE_FORK))
			sync_entity_load_avg(&p->se);
M
Mike Galbraith 已提交
6662

6663
		new_cpu = find_idlest_cpu(sd, p, cpu, prev_cpu, sd_flag);
6664 6665 6666 6667 6668 6669 6670
	} else if (sd_flag & SD_BALANCE_WAKE) { /* XXX always ? */
		/* Fast path */

		new_cpu = select_idle_sibling(p, prev_cpu, new_cpu);

		if (want_affine)
			current->recent_used_cpu = cpu;
6671
	}
6672
	rcu_read_unlock();
6673

6674
	return new_cpu;
6675
}
6676

6677 6678
static void detach_entity_cfs_rq(struct sched_entity *se);

6679
/*
6680
 * Called immediately before a task is migrated to a new CPU; task_cpu(p) and
6681
 * cfs_rq_of(p) references at time of call are still valid and identify the
6682
 * previous CPU. The caller guarantees p->pi_lock or task_rq(p)->lock is held.
6683
 */
6684
static void migrate_task_rq_fair(struct task_struct *p)
6685
{
6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711
	/*
	 * As blocked tasks retain absolute vruntime the migration needs to
	 * deal with this by subtracting the old and adding the new
	 * min_vruntime -- the latter is done by enqueue_entity() when placing
	 * the task on the new runqueue.
	 */
	if (p->state == TASK_WAKING) {
		struct sched_entity *se = &p->se;
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
		u64 min_vruntime;

#ifndef CONFIG_64BIT
		u64 min_vruntime_copy;

		do {
			min_vruntime_copy = cfs_rq->min_vruntime_copy;
			smp_rmb();
			min_vruntime = cfs_rq->min_vruntime;
		} while (min_vruntime != min_vruntime_copy);
#else
		min_vruntime = cfs_rq->min_vruntime;
#endif

		se->vruntime -= min_vruntime;
	}

6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730
	if (p->on_rq == TASK_ON_RQ_MIGRATING) {
		/*
		 * In case of TASK_ON_RQ_MIGRATING we in fact hold the 'old'
		 * rq->lock and can modify state directly.
		 */
		lockdep_assert_held(&task_rq(p)->lock);
		detach_entity_cfs_rq(&p->se);

	} else {
		/*
		 * We are supposed to update the task to "current" time, then
		 * its up to date and ready to go to new CPU/cfs_rq. But we
		 * have difficulty in getting what current time is, so simply
		 * throw away the out-of-date time. This will result in the
		 * wakee task is less decayed, but giving the wakee more load
		 * sounds not bad.
		 */
		remove_entity_load_avg(&p->se);
	}
6731 6732 6733

	/* Tell new CPU we are migrated */
	p->se.avg.last_update_time = 0;
6734 6735

	/* We have migrated, no longer consider this task hot */
6736
	p->se.exec_start = 0;
6737
}
6738 6739 6740 6741 6742

static void task_dead_fair(struct task_struct *p)
{
	remove_entity_load_avg(&p->se);
}
6743 6744
#endif /* CONFIG_SMP */

6745
static unsigned long wakeup_gran(struct sched_entity *se)
6746 6747 6748 6749
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
6750 6751
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
6752 6753 6754 6755 6756 6757 6758 6759 6760
	 *
	 * By using 'se' instead of 'curr' we penalize light tasks, so
	 * they get preempted easier. That is, if 'se' < 'curr' then
	 * the resulting gran will be larger, therefore penalizing the
	 * lighter, if otoh 'se' > 'curr' then the resulting gran will
	 * be smaller, again penalizing the lighter task.
	 *
	 * This is especially important for buddies when the leftmost
	 * task is higher priority than the buddy.
6761
	 */
6762
	return calc_delta_fair(gran, se);
6763 6764
}

6765 6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786
/*
 * Should 'se' preempt 'curr'.
 *
 *             |s1
 *        |s2
 *   |s3
 *         g
 *      |<--->|c
 *
 *  w(c, s1) = -1
 *  w(c, s2) =  0
 *  w(c, s3) =  1
 *
 */
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se)
{
	s64 gran, vdiff = curr->vruntime - se->vruntime;

	if (vdiff <= 0)
		return -1;

6787
	gran = wakeup_gran(se);
6788 6789 6790 6791 6792 6793
	if (vdiff > gran)
		return 1;

	return 0;
}

6794 6795
static void set_last_buddy(struct sched_entity *se)
{
6796 6797 6798
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

6799 6800 6801
	for_each_sched_entity(se) {
		if (SCHED_WARN_ON(!se->on_rq))
			return;
6802
		cfs_rq_of(se)->last = se;
6803
	}
6804 6805 6806 6807
}

static void set_next_buddy(struct sched_entity *se)
{
6808 6809 6810
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

6811 6812 6813
	for_each_sched_entity(se) {
		if (SCHED_WARN_ON(!se->on_rq))
			return;
6814
		cfs_rq_of(se)->next = se;
6815
	}
6816 6817
}

6818 6819
static void set_skip_buddy(struct sched_entity *se)
{
6820 6821
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
6822 6823
}

6824 6825 6826
/*
 * Preempt the current task with a newly woken task if needed:
 */
6827
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
6828 6829
{
	struct task_struct *curr = rq->curr;
6830
	struct sched_entity *se = &curr->se, *pse = &p->se;
6831
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
6832
	int scale = cfs_rq->nr_running >= sched_nr_latency;
6833
	int next_buddy_marked = 0;
6834

I
Ingo Molnar 已提交
6835 6836 6837
	if (unlikely(se == pse))
		return;

6838
	/*
6839
	 * This is possible from callers such as attach_tasks(), in which we
6840 6841 6842 6843 6844 6845 6846
	 * unconditionally check_prempt_curr() after an enqueue (which may have
	 * lead to a throttle).  This both saves work and prevents false
	 * next-buddy nomination below.
	 */
	if (unlikely(throttled_hierarchy(cfs_rq_of(pse))))
		return;

6847
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
6848
		set_next_buddy(pse);
6849 6850
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
6851

6852 6853 6854
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
6855 6856 6857 6858 6859 6860
	 *
	 * Note: this also catches the edge-case of curr being in a throttled
	 * group (e.g. via set_curr_task), since update_curr() (in the
	 * enqueue of curr) will have resulted in resched being set.  This
	 * prevents us from potentially nominating it as a false LAST_BUDDY
	 * below.
6861 6862 6863 6864
	 */
	if (test_tsk_need_resched(curr))
		return;

6865 6866 6867 6868 6869
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

6870
	/*
6871 6872
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
6873
	 */
6874
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
6875
		return;
6876

6877
	find_matching_se(&se, &pse);
6878
	update_curr(cfs_rq_of(se));
6879
	BUG_ON(!pse);
6880 6881 6882 6883 6884 6885 6886
	if (wakeup_preempt_entity(se, pse) == 1) {
		/*
		 * Bias pick_next to pick the sched entity that is
		 * triggering this preemption.
		 */
		if (!next_buddy_marked)
			set_next_buddy(pse);
6887
		goto preempt;
6888
	}
6889

6890
	return;
6891

6892
preempt:
6893
	resched_curr(rq);
6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907
	/*
	 * Only set the backward buddy when the current task is still
	 * on the rq. This can happen when a wakeup gets interleaved
	 * with schedule on the ->pre_schedule() or idle_balance()
	 * point, either of which can * drop the rq lock.
	 *
	 * Also, during early boot the idle thread is in the fair class,
	 * for obvious reasons its a bad idea to schedule back to it.
	 */
	if (unlikely(!se->on_rq || curr == rq->idle))
		return;

	if (sched_feat(LAST_BUDDY) && scale && entity_is_task(se))
		set_last_buddy(se);
6908 6909
}

6910
static struct task_struct *
6911
pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
6912 6913 6914
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
6915
	struct task_struct *p;
6916
	int new_tasks;
6917

6918
again:
6919
	if (!cfs_rq->nr_running)
6920
		goto idle;
6921

6922
#ifdef CONFIG_FAIR_GROUP_SCHED
6923
	if (prev->sched_class != &fair_sched_class)
6924 6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935 6936 6937 6938 6939 6940 6941 6942
		goto simple;

	/*
	 * Because of the set_next_buddy() in dequeue_task_fair() it is rather
	 * likely that a next task is from the same cgroup as the current.
	 *
	 * Therefore attempt to avoid putting and setting the entire cgroup
	 * hierarchy, only change the part that actually changes.
	 */

	do {
		struct sched_entity *curr = cfs_rq->curr;

		/*
		 * Since we got here without doing put_prev_entity() we also
		 * have to consider cfs_rq->curr. If it is still a runnable
		 * entity, update_curr() will update its vruntime, otherwise
		 * forget we've ever seen it.
		 */
6943 6944 6945 6946 6947
		if (curr) {
			if (curr->on_rq)
				update_curr(cfs_rq);
			else
				curr = NULL;
6948

6949 6950 6951
			/*
			 * This call to check_cfs_rq_runtime() will do the
			 * throttle and dequeue its entity in the parent(s).
6952
			 * Therefore the nr_running test will indeed
6953 6954
			 * be correct.
			 */
6955 6956 6957 6958 6959 6960
			if (unlikely(check_cfs_rq_runtime(cfs_rq))) {
				cfs_rq = &rq->cfs;

				if (!cfs_rq->nr_running)
					goto idle;

6961
				goto simple;
6962
			}
6963
		}
6964 6965 6966 6967 6968 6969 6970 6971 6972 6973 6974 6975 6976 6977 6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996

		se = pick_next_entity(cfs_rq, curr);
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

	p = task_of(se);

	/*
	 * Since we haven't yet done put_prev_entity and if the selected task
	 * is a different task than we started out with, try and touch the
	 * least amount of cfs_rqs.
	 */
	if (prev != p) {
		struct sched_entity *pse = &prev->se;

		while (!(cfs_rq = is_same_group(se, pse))) {
			int se_depth = se->depth;
			int pse_depth = pse->depth;

			if (se_depth <= pse_depth) {
				put_prev_entity(cfs_rq_of(pse), pse);
				pse = parent_entity(pse);
			}
			if (se_depth >= pse_depth) {
				set_next_entity(cfs_rq_of(se), se);
				se = parent_entity(se);
			}
		}

		put_prev_entity(cfs_rq, pse);
		set_next_entity(cfs_rq, se);
	}

6997
	goto done;
6998 6999
simple:
#endif
7000

7001
	put_prev_task(rq, prev);
7002

7003
	do {
7004
		se = pick_next_entity(cfs_rq, NULL);
7005
		set_next_entity(cfs_rq, se);
7006 7007 7008
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
7009
	p = task_of(se);
7010

7011
done: __maybe_unused;
7012 7013 7014 7015 7016 7017 7018 7019 7020
#ifdef CONFIG_SMP
	/*
	 * Move the next running task to the front of
	 * the list, so our cfs_tasks list becomes MRU
	 * one.
	 */
	list_move(&p->se.group_node, &rq->cfs_tasks);
#endif

7021 7022
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
7023 7024

	return p;
7025 7026

idle:
7027 7028
	new_tasks = idle_balance(rq, rf);

7029 7030 7031 7032 7033
	/*
	 * Because idle_balance() releases (and re-acquires) rq->lock, it is
	 * possible for any higher priority task to appear. In that case we
	 * must re-start the pick_next_entity() loop.
	 */
7034
	if (new_tasks < 0)
7035 7036
		return RETRY_TASK;

7037
	if (new_tasks > 0)
7038 7039 7040
		goto again;

	return NULL;
7041 7042 7043 7044 7045
}

/*
 * Account for a descheduled task:
 */
7046
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
7047 7048 7049 7050 7051 7052
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
7053
		put_prev_entity(cfs_rq, se);
7054 7055 7056
	}
}

7057 7058 7059 7060 7061 7062 7063 7064 7065 7066 7067 7068 7069 7070 7071 7072 7073 7074 7075 7076 7077 7078 7079 7080 7081
/*
 * sched_yield() is very simple
 *
 * The magic of dealing with the ->skip buddy is in pick_next_entity.
 */
static void yield_task_fair(struct rq *rq)
{
	struct task_struct *curr = rq->curr;
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
	struct sched_entity *se = &curr->se;

	/*
	 * Are we the only task in the tree?
	 */
	if (unlikely(rq->nr_running == 1))
		return;

	clear_buddies(cfs_rq, se);

	if (curr->policy != SCHED_BATCH) {
		update_rq_clock(rq);
		/*
		 * Update run-time statistics of the 'current'.
		 */
		update_curr(cfs_rq);
7082 7083 7084 7085 7086
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
7087
		rq_clock_skip_update(rq);
7088 7089 7090 7091 7092
	}

	set_skip_buddy(se);
}

7093 7094 7095 7096
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

7097 7098
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
7099 7100 7101 7102 7103 7104 7105 7106 7107 7108
		return false;

	/* Tell the scheduler that we'd really like pse to run next. */
	set_next_buddy(se);

	yield_task_fair(rq);

	return true;
}

7109
#ifdef CONFIG_SMP
7110
/**************************************************
P
Peter Zijlstra 已提交
7111 7112 7113 7114 7115
 * Fair scheduling class load-balancing methods.
 *
 * BASICS
 *
 * The purpose of load-balancing is to achieve the same basic fairness the
7116
 * per-CPU scheduler provides, namely provide a proportional amount of compute
P
Peter Zijlstra 已提交
7117 7118 7119 7120
 * time to each task. This is expressed in the following equation:
 *
 *   W_i,n/P_i == W_j,n/P_j for all i,j                               (1)
 *
7121
 * Where W_i,n is the n-th weight average for CPU i. The instantaneous weight
P
Peter Zijlstra 已提交
7122 7123 7124 7125
 * W_i,0 is defined as:
 *
 *   W_i,0 = \Sum_j w_i,j                                             (2)
 *
7126
 * Where w_i,j is the weight of the j-th runnable task on CPU i. This weight
7127
 * is derived from the nice value as per sched_prio_to_weight[].
P
Peter Zijlstra 已提交
7128 7129 7130 7131 7132 7133
 *
 * The weight average is an exponential decay average of the instantaneous
 * weight:
 *
 *   W'_i,n = (2^n - 1) / 2^n * W_i,n + 1 / 2^n * W_i,0               (3)
 *
7134
 * C_i is the compute capacity of CPU i, typically it is the
P
Peter Zijlstra 已提交
7135 7136 7137 7138 7139 7140
 * fraction of 'recent' time available for SCHED_OTHER task execution. But it
 * can also include other factors [XXX].
 *
 * To achieve this balance we define a measure of imbalance which follows
 * directly from (1):
 *
7141
 *   imb_i,j = max{ avg(W/C), W_i/C_i } - min{ avg(W/C), W_j/C_j }    (4)
P
Peter Zijlstra 已提交
7142 7143 7144 7145 7146 7147 7148 7149 7150 7151 7152 7153 7154
 *
 * We them move tasks around to minimize the imbalance. In the continuous
 * function space it is obvious this converges, in the discrete case we get
 * a few fun cases generally called infeasible weight scenarios.
 *
 * [XXX expand on:
 *     - infeasible weights;
 *     - local vs global optima in the discrete case. ]
 *
 *
 * SCHED DOMAINS
 *
 * In order to solve the imbalance equation (4), and avoid the obvious O(n^2)
7155
 * for all i,j solution, we create a tree of CPUs that follows the hardware
P
Peter Zijlstra 已提交
7156
 * topology where each level pairs two lower groups (or better). This results
7157
 * in O(log n) layers. Furthermore we reduce the number of CPUs going up the
P
Peter Zijlstra 已提交
7158
 * tree to only the first of the previous level and we decrease the frequency
7159
 * of load-balance at each level inv. proportional to the number of CPUs in
P
Peter Zijlstra 已提交
7160 7161 7162 7163 7164 7165 7166 7167
 * the groups.
 *
 * This yields:
 *
 *     log_2 n     1     n
 *   \Sum       { --- * --- * 2^i } = O(n)                            (5)
 *     i = 0      2^i   2^i
 *                               `- size of each group
7168
 *         |         |     `- number of CPUs doing load-balance
P
Peter Zijlstra 已提交
7169 7170 7171 7172 7173 7174 7175
 *         |         `- freq
 *         `- sum over all levels
 *
 * Coupled with a limit on how many tasks we can migrate every balance pass,
 * this makes (5) the runtime complexity of the balancer.
 *
 * An important property here is that each CPU is still (indirectly) connected
7176
 * to every other CPU in at most O(log n) steps:
P
Peter Zijlstra 已提交
7177 7178 7179
 *
 * The adjacency matrix of the resulting graph is given by:
 *
7180
 *             log_2 n
P
Peter Zijlstra 已提交
7181 7182 7183 7184 7185 7186 7187
 *   A_i,j = \Union     (i % 2^k == 0) && i / 2^(k+1) == j / 2^(k+1)  (6)
 *             k = 0
 *
 * And you'll find that:
 *
 *   A^(log_2 n)_i,j != 0  for all i,j                                (7)
 *
7188
 * Showing there's indeed a path between every CPU in at most O(log n) steps.
P
Peter Zijlstra 已提交
7189 7190 7191 7192 7193 7194 7195 7196 7197
 * The task movement gives a factor of O(m), giving a convergence complexity
 * of:
 *
 *   O(nm log n),  n := nr_cpus, m := nr_tasks                        (8)
 *
 *
 * WORK CONSERVING
 *
 * In order to avoid CPUs going idle while there's still work to do, new idle
7198
 * balancing is more aggressive and has the newly idle CPU iterate up the domain
P
Peter Zijlstra 已提交
7199 7200 7201 7202 7203 7204 7205 7206 7207 7208 7209 7210 7211 7212 7213 7214 7215 7216 7217 7218
 * tree itself instead of relying on other CPUs to bring it work.
 *
 * This adds some complexity to both (5) and (8) but it reduces the total idle
 * time.
 *
 * [XXX more?]
 *
 *
 * CGROUPS
 *
 * Cgroups make a horror show out of (2), instead of a simple sum we get:
 *
 *                                s_k,i
 *   W_i,0 = \Sum_j \Prod_k w_k * -----                               (9)
 *                                 S_k
 *
 * Where
 *
 *   s_k,i = \Sum_j w_i,j,k  and  S_k = \Sum_i s_k,i                 (10)
 *
7219
 * w_i,j,k is the weight of the j-th runnable task in the k-th cgroup on CPU i.
P
Peter Zijlstra 已提交
7220 7221 7222 7223 7224 7225
 *
 * The big problem is S_k, its a global sum needed to compute a local (W_i)
 * property.
 *
 * [XXX write more on how we solve this.. _after_ merging pjt's patches that
 *      rewrite all of this once again.]
7226
 */
7227

7228 7229
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

7230 7231
enum fbq_type { regular, remote, all };

7232
#define LBF_ALL_PINNED	0x01
7233
#define LBF_NEED_BREAK	0x02
7234 7235
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
7236
#define LBF_NOHZ_STATS	0x10
7237
#define LBF_NOHZ_AGAIN	0x20
7238 7239 7240 7241 7242

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
7243
	int			src_cpu;
7244 7245 7246 7247

	int			dst_cpu;
	struct rq		*dst_rq;

7248 7249
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
7250
	enum cpu_idle_type	idle;
7251
	long			imbalance;
7252 7253 7254
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

7255
	unsigned int		flags;
7256 7257 7258 7259

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
7260 7261

	enum fbq_type		fbq_type;
7262
	struct list_head	tasks;
7263 7264
};

7265 7266 7267
/*
 * Is this task likely cache-hot:
 */
7268
static int task_hot(struct task_struct *p, struct lb_env *env)
7269 7270 7271
{
	s64 delta;

7272 7273
	lockdep_assert_held(&env->src_rq->lock);

7274 7275 7276 7277 7278 7279 7280 7281 7282
	if (p->sched_class != &fair_sched_class)
		return 0;

	if (unlikely(p->policy == SCHED_IDLE))
		return 0;

	/*
	 * Buddy candidates are cache hot:
	 */
7283
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
7284 7285 7286 7287 7288 7289 7290 7291 7292
			(&p->se == cfs_rq_of(&p->se)->next ||
			 &p->se == cfs_rq_of(&p->se)->last))
		return 1;

	if (sysctl_sched_migration_cost == -1)
		return 1;
	if (sysctl_sched_migration_cost == 0)
		return 0;

7293
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
7294 7295 7296 7297

	return delta < (s64)sysctl_sched_migration_cost;
}

7298
#ifdef CONFIG_NUMA_BALANCING
7299
/*
7300 7301 7302
 * Returns 1, if task migration degrades locality
 * Returns 0, if task migration improves locality i.e migration preferred.
 * Returns -1, if task migration is not affected by locality.
7303
 */
7304
static int migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
7305
{
7306
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
7307
	unsigned long src_faults, dst_faults;
7308 7309
	int src_nid, dst_nid;

7310
	if (!static_branch_likely(&sched_numa_balancing))
7311 7312
		return -1;

7313
	if (!p->numa_faults || !(env->sd->flags & SD_NUMA))
7314
		return -1;
7315 7316 7317 7318

	src_nid = cpu_to_node(env->src_cpu);
	dst_nid = cpu_to_node(env->dst_cpu);

7319
	if (src_nid == dst_nid)
7320
		return -1;
7321

7322 7323 7324 7325 7326 7327 7328
	/* Migrating away from the preferred node is always bad. */
	if (src_nid == p->numa_preferred_nid) {
		if (env->src_rq->nr_running > env->src_rq->nr_preferred_running)
			return 1;
		else
			return -1;
	}
7329

7330 7331
	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
7332
		return 0;
7333

7334 7335 7336 7337
	/* Leaving a core idle is often worse than degrading locality. */
	if (env->idle != CPU_NOT_IDLE)
		return -1;

7338 7339 7340 7341 7342 7343
	if (numa_group) {
		src_faults = group_faults(p, src_nid);
		dst_faults = group_faults(p, dst_nid);
	} else {
		src_faults = task_faults(p, src_nid);
		dst_faults = task_faults(p, dst_nid);
7344 7345
	}

7346
	return dst_faults < src_faults;
7347 7348
}

7349
#else
7350
static inline int migrate_degrades_locality(struct task_struct *p,
7351 7352
					     struct lb_env *env)
{
7353
	return -1;
7354
}
7355 7356
#endif

7357 7358 7359 7360
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
7361
int can_migrate_task(struct task_struct *p, struct lb_env *env)
7362
{
7363
	int tsk_cache_hot;
7364 7365 7366

	lockdep_assert_held(&env->src_rq->lock);

7367 7368
	/*
	 * We do not migrate tasks that are:
7369
	 * 1) throttled_lb_pair, or
7370
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
7371 7372
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
7373
	 */
7374 7375 7376
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

7377
	if (!cpumask_test_cpu(env->dst_cpu, &p->cpus_allowed)) {
7378
		int cpu;
7379

7380
		schedstat_inc(p->se.statistics.nr_failed_migrations_affine);
7381

7382 7383
		env->flags |= LBF_SOME_PINNED;

7384
		/*
7385
		 * Remember if this task can be migrated to any other CPU in
7386 7387 7388
		 * our sched_group. We may want to revisit it if we couldn't
		 * meet load balance goals by pulling other tasks on src_cpu.
		 *
7389 7390
		 * Avoid computing new_dst_cpu for NEWLY_IDLE or if we have
		 * already computed one in current iteration.
7391
		 */
7392
		if (env->idle == CPU_NEWLY_IDLE || (env->flags & LBF_DST_PINNED))
7393 7394
			return 0;

7395
		/* Prevent to re-select dst_cpu via env's CPUs: */
7396
		for_each_cpu_and(cpu, env->dst_grpmask, env->cpus) {
7397
			if (cpumask_test_cpu(cpu, &p->cpus_allowed)) {
7398
				env->flags |= LBF_DST_PINNED;
7399 7400 7401
				env->new_dst_cpu = cpu;
				break;
			}
7402
		}
7403

7404 7405
		return 0;
	}
7406 7407

	/* Record that we found atleast one task that could run on dst_cpu */
7408
	env->flags &= ~LBF_ALL_PINNED;
7409

7410
	if (task_running(env->src_rq, p)) {
7411
		schedstat_inc(p->se.statistics.nr_failed_migrations_running);
7412 7413 7414 7415 7416
		return 0;
	}

	/*
	 * Aggressive migration if:
7417 7418 7419
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
7420
	 */
7421 7422 7423
	tsk_cache_hot = migrate_degrades_locality(p, env);
	if (tsk_cache_hot == -1)
		tsk_cache_hot = task_hot(p, env);
7424

7425
	if (tsk_cache_hot <= 0 ||
7426
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
7427
		if (tsk_cache_hot == 1) {
7428 7429
			schedstat_inc(env->sd->lb_hot_gained[env->idle]);
			schedstat_inc(p->se.statistics.nr_forced_migrations);
7430
		}
7431 7432 7433
		return 1;
	}

7434
	schedstat_inc(p->se.statistics.nr_failed_migrations_hot);
Z
Zhang Hang 已提交
7435
	return 0;
7436 7437
}

7438
/*
7439 7440 7441 7442 7443 7444 7445
 * detach_task() -- detach the task for the migration specified in env
 */
static void detach_task(struct task_struct *p, struct lb_env *env)
{
	lockdep_assert_held(&env->src_rq->lock);

	p->on_rq = TASK_ON_RQ_MIGRATING;
7446
	deactivate_task(env->src_rq, p, DEQUEUE_NOCLOCK);
7447 7448 7449
	set_task_cpu(p, env->dst_cpu);
}

7450
/*
7451
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
7452 7453
 * part of active balancing operations within "domain".
 *
7454
 * Returns a task if successful and NULL otherwise.
7455
 */
7456
static struct task_struct *detach_one_task(struct lb_env *env)
7457
{
7458
	struct task_struct *p;
7459

7460 7461
	lockdep_assert_held(&env->src_rq->lock);

7462 7463
	list_for_each_entry_reverse(p,
			&env->src_rq->cfs_tasks, se.group_node) {
7464 7465
		if (!can_migrate_task(p, env))
			continue;
7466

7467
		detach_task(p, env);
7468

7469
		/*
7470
		 * Right now, this is only the second place where
7471
		 * lb_gained[env->idle] is updated (other is detach_tasks)
7472
		 * so we can safely collect stats here rather than
7473
		 * inside detach_tasks().
7474
		 */
7475
		schedstat_inc(env->sd->lb_gained[env->idle]);
7476
		return p;
7477
	}
7478
	return NULL;
7479 7480
}

7481 7482
static const unsigned int sched_nr_migrate_break = 32;

7483
/*
7484 7485
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
7486
 *
7487
 * Returns number of detached tasks if successful and 0 otherwise.
7488
 */
7489
static int detach_tasks(struct lb_env *env)
7490
{
7491 7492
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
7493
	unsigned long load;
7494 7495 7496
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
7497

7498
	if (env->imbalance <= 0)
7499
		return 0;
7500

7501
	while (!list_empty(tasks)) {
7502 7503 7504 7505 7506 7507 7508
		/*
		 * We don't want to steal all, otherwise we may be treated likewise,
		 * which could at worst lead to a livelock crash.
		 */
		if (env->idle != CPU_NOT_IDLE && env->src_rq->nr_running <= 1)
			break;

7509
		p = list_last_entry(tasks, struct task_struct, se.group_node);
7510

7511 7512
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
7513
		if (env->loop > env->loop_max)
7514
			break;
7515 7516

		/* take a breather every nr_migrate tasks */
7517
		if (env->loop > env->loop_break) {
7518
			env->loop_break += sched_nr_migrate_break;
7519
			env->flags |= LBF_NEED_BREAK;
7520
			break;
7521
		}
7522

7523
		if (!can_migrate_task(p, env))
7524 7525 7526
			goto next;

		load = task_h_load(p);
7527

7528
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
7529 7530
			goto next;

7531
		if ((load / 2) > env->imbalance)
7532
			goto next;
7533

7534 7535 7536 7537
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
7538
		env->imbalance -= load;
7539 7540

#ifdef CONFIG_PREEMPT
7541 7542
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
7543
		 * kernels will stop after the first task is detached to minimize
7544 7545
		 * the critical section.
		 */
7546
		if (env->idle == CPU_NEWLY_IDLE)
7547
			break;
7548 7549
#endif

7550 7551 7552 7553
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
7554
		if (env->imbalance <= 0)
7555
			break;
7556 7557 7558

		continue;
next:
7559
		list_move(&p->se.group_node, tasks);
7560
	}
7561

7562
	/*
7563 7564 7565
	 * Right now, this is one of only two places we collect this stat
	 * so we can safely collect detach_one_task() stats here rather
	 * than inside detach_one_task().
7566
	 */
7567
	schedstat_add(env->sd->lb_gained[env->idle], detached);
7568

7569 7570 7571 7572 7573 7574 7575 7576 7577 7578 7579
	return detached;
}

/*
 * attach_task() -- attach the task detached by detach_task() to its new rq.
 */
static void attach_task(struct rq *rq, struct task_struct *p)
{
	lockdep_assert_held(&rq->lock);

	BUG_ON(task_rq(p) != rq);
7580
	activate_task(rq, p, ENQUEUE_NOCLOCK);
7581
	p->on_rq = TASK_ON_RQ_QUEUED;
7582 7583 7584 7585 7586 7587 7588 7589 7590
	check_preempt_curr(rq, p, 0);
}

/*
 * attach_one_task() -- attaches the task returned from detach_one_task() to
 * its new rq.
 */
static void attach_one_task(struct rq *rq, struct task_struct *p)
{
7591 7592 7593
	struct rq_flags rf;

	rq_lock(rq, &rf);
7594
	update_rq_clock(rq);
7595
	attach_task(rq, p);
7596
	rq_unlock(rq, &rf);
7597 7598 7599 7600 7601 7602 7603 7604 7605 7606
}

/*
 * attach_tasks() -- attaches all tasks detached by detach_tasks() to their
 * new rq.
 */
static void attach_tasks(struct lb_env *env)
{
	struct list_head *tasks = &env->tasks;
	struct task_struct *p;
7607
	struct rq_flags rf;
7608

7609
	rq_lock(env->dst_rq, &rf);
7610
	update_rq_clock(env->dst_rq);
7611 7612 7613 7614

	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
		list_del_init(&p->se.group_node);
7615

7616 7617 7618
		attach_task(env->dst_rq, p);
	}

7619
	rq_unlock(env->dst_rq, &rf);
7620 7621
}

7622 7623 7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634
static inline bool cfs_rq_has_blocked(struct cfs_rq *cfs_rq)
{
	if (cfs_rq->avg.load_avg)
		return true;

	if (cfs_rq->avg.util_avg)
		return true;

	return false;
}

#ifdef CONFIG_FAIR_GROUP_SCHED

7635 7636 7637 7638 7639 7640 7641 7642 7643 7644 7645
static inline bool cfs_rq_is_decayed(struct cfs_rq *cfs_rq)
{
	if (cfs_rq->load.weight)
		return false;

	if (cfs_rq->avg.load_sum)
		return false;

	if (cfs_rq->avg.util_sum)
		return false;

7646
	if (cfs_rq->avg.runnable_load_sum)
7647 7648 7649 7650 7651
		return false;

	return true;
}

7652
static void update_blocked_averages(int cpu)
7653 7654
{
	struct rq *rq = cpu_rq(cpu);
7655
	struct cfs_rq *cfs_rq, *pos;
7656
	struct rq_flags rf;
7657
	bool done = true;
7658

7659
	rq_lock_irqsave(rq, &rf);
7660
	update_rq_clock(rq);
7661

7662 7663 7664 7665
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
7666
	for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos) {
7667 7668
		struct sched_entity *se;

7669 7670 7671
		/* throttled entities do not contribute to load */
		if (throttled_hierarchy(cfs_rq))
			continue;
7672

7673
		if (update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq))
7674
			update_tg_load_avg(cfs_rq, 0);
7675

7676 7677 7678
		/* Propagate pending load changes to the parent, if any: */
		se = cfs_rq->tg->se[cpu];
		if (se && !skip_blocked_update(se))
7679
			update_load_avg(cfs_rq_of(se), se, 0);
7680 7681 7682 7683 7684 7685 7686

		/*
		 * There can be a lot of idle CPU cgroups.  Don't let fully
		 * decayed cfs_rqs linger on the list.
		 */
		if (cfs_rq_is_decayed(cfs_rq))
			list_del_leaf_cfs_rq(cfs_rq);
7687 7688 7689

		/* Don't need periodic decay once load/util_avg are null */
		if (cfs_rq_has_blocked(cfs_rq))
7690
			done = false;
7691
	}
7692 7693 7694

#ifdef CONFIG_NO_HZ_COMMON
	rq->last_blocked_load_update_tick = jiffies;
7695 7696
	if (done)
		rq->has_blocked_load = 0;
7697
#endif
7698
	rq_unlock_irqrestore(rq, &rf);
7699 7700
}

7701
/*
7702
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
7703 7704 7705
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
7706
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
7707
{
7708 7709
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
7710
	unsigned long now = jiffies;
7711
	unsigned long load;
7712

7713
	if (cfs_rq->last_h_load_update == now)
7714 7715
		return;

7716 7717 7718 7719 7720 7721 7722
	cfs_rq->h_load_next = NULL;
	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
		cfs_rq->h_load_next = se;
		if (cfs_rq->last_h_load_update == now)
			break;
	}
7723

7724
	if (!se) {
7725
		cfs_rq->h_load = cfs_rq_load_avg(cfs_rq);
7726 7727 7728 7729 7730
		cfs_rq->last_h_load_update = now;
	}

	while ((se = cfs_rq->h_load_next) != NULL) {
		load = cfs_rq->h_load;
7731 7732
		load = div64_ul(load * se->avg.load_avg,
			cfs_rq_load_avg(cfs_rq) + 1);
7733 7734 7735 7736
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
7737 7738
}

7739
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
7740
{
7741
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
7742

7743
	update_cfs_rq_h_load(cfs_rq);
7744
	return div64_ul(p->se.avg.load_avg * cfs_rq->h_load,
7745
			cfs_rq_load_avg(cfs_rq) + 1);
P
Peter Zijlstra 已提交
7746 7747
}
#else
7748
static inline void update_blocked_averages(int cpu)
7749
{
7750 7751
	struct rq *rq = cpu_rq(cpu);
	struct cfs_rq *cfs_rq = &rq->cfs;
7752
	struct rq_flags rf;
7753

7754
	rq_lock_irqsave(rq, &rf);
7755
	update_rq_clock(rq);
7756
	update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq);
7757 7758
#ifdef CONFIG_NO_HZ_COMMON
	rq->last_blocked_load_update_tick = jiffies;
7759
	if (!cfs_rq_has_blocked(cfs_rq))
7760
		rq->has_blocked_load = 0;
7761
#endif
7762
	rq_unlock_irqrestore(rq, &rf);
7763 7764
}

7765
static unsigned long task_h_load(struct task_struct *p)
7766
{
7767
	return p->se.avg.load_avg;
7768
}
P
Peter Zijlstra 已提交
7769
#endif
7770 7771

/********** Helpers for find_busiest_group ************************/
7772 7773 7774 7775 7776 7777 7778

enum group_type {
	group_other = 0,
	group_imbalanced,
	group_overloaded,
};

7779 7780 7781 7782 7783 7784 7785
/*
 * sg_lb_stats - stats of a sched_group required for load_balancing
 */
struct sg_lb_stats {
	unsigned long avg_load; /*Avg load across the CPUs of the group */
	unsigned long group_load; /* Total load over the CPUs of the group */
	unsigned long sum_weighted_load; /* Weighted load of group's tasks */
J
Joonsoo Kim 已提交
7786
	unsigned long load_per_task;
7787
	unsigned long group_capacity;
7788
	unsigned long group_util; /* Total utilization of the group */
7789 7790 7791
	unsigned int sum_nr_running; /* Nr tasks running in the group */
	unsigned int idle_cpus;
	unsigned int group_weight;
7792
	enum group_type group_type;
7793
	int group_no_capacity;
7794 7795 7796 7797
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
7798 7799
};

J
Joonsoo Kim 已提交
7800 7801 7802 7803 7804 7805 7806
/*
 * sd_lb_stats - Structure to store the statistics of a sched_domain
 *		 during load balancing.
 */
struct sd_lb_stats {
	struct sched_group *busiest;	/* Busiest group in this sd */
	struct sched_group *local;	/* Local group in this sd */
7807
	unsigned long total_running;
J
Joonsoo Kim 已提交
7808
	unsigned long total_load;	/* Total load of all groups in sd */
7809
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
7810 7811 7812
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
7813
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
7814 7815
};

7816 7817 7818 7819 7820 7821 7822 7823 7824 7825 7826
static inline void init_sd_lb_stats(struct sd_lb_stats *sds)
{
	/*
	 * Skimp on the clearing to avoid duplicate work. We can avoid clearing
	 * local_stat because update_sg_lb_stats() does a full clear/assignment.
	 * We must however clear busiest_stat::avg_load because
	 * update_sd_pick_busiest() reads this before assignment.
	 */
	*sds = (struct sd_lb_stats){
		.busiest = NULL,
		.local = NULL,
7827
		.total_running = 0UL,
7828
		.total_load = 0UL,
7829
		.total_capacity = 0UL,
7830 7831
		.busiest_stat = {
			.avg_load = 0UL,
7832 7833
			.sum_nr_running = 0,
			.group_type = group_other,
7834 7835 7836 7837
		},
	};
}

7838 7839 7840
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
7841
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
7842 7843
 *
 * Return: The load index.
7844 7845 7846 7847 7848 7849 7850 7851 7852 7853 7854 7855 7856 7857 7858 7859 7860 7861 7862 7863 7864 7865
 */
static inline int get_sd_load_idx(struct sched_domain *sd,
					enum cpu_idle_type idle)
{
	int load_idx;

	switch (idle) {
	case CPU_NOT_IDLE:
		load_idx = sd->busy_idx;
		break;

	case CPU_NEWLY_IDLE:
		load_idx = sd->newidle_idx;
		break;
	default:
		load_idx = sd->idle_idx;
		break;
	}

	return load_idx;
}

7866
static unsigned long scale_rt_capacity(int cpu)
7867 7868
{
	struct rq *rq = cpu_rq(cpu);
7869
	u64 total, used, age_stamp, avg;
7870
	s64 delta;
7871

7872 7873 7874 7875
	/*
	 * Since we're reading these variables without serialization make sure
	 * we read them once before doing sanity checks on them.
	 */
7876 7877
	age_stamp = READ_ONCE(rq->age_stamp);
	avg = READ_ONCE(rq->rt_avg);
7878
	delta = __rq_clock_broken(rq) - age_stamp;
7879

7880 7881 7882 7883
	if (unlikely(delta < 0))
		delta = 0;

	total = sched_avg_period() + delta;
7884

7885
	used = div_u64(avg, total);
7886

7887 7888
	if (likely(used < SCHED_CAPACITY_SCALE))
		return SCHED_CAPACITY_SCALE - used;
7889

7890
	return 1;
7891 7892
}

7893
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
7894
{
7895
	unsigned long capacity = arch_scale_cpu_capacity(sd, cpu);
7896 7897
	struct sched_group *sdg = sd->groups;

7898
	cpu_rq(cpu)->cpu_capacity_orig = capacity;
7899

7900
	capacity *= scale_rt_capacity(cpu);
7901
	capacity >>= SCHED_CAPACITY_SHIFT;
7902

7903 7904
	if (!capacity)
		capacity = 1;
7905

7906 7907
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
7908
	sdg->sgc->min_capacity = capacity;
7909 7910
}

7911
void update_group_capacity(struct sched_domain *sd, int cpu)
7912 7913 7914
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
7915
	unsigned long capacity, min_capacity;
7916 7917 7918 7919
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
7920
	sdg->sgc->next_update = jiffies + interval;
7921 7922

	if (!child) {
7923
		update_cpu_capacity(sd, cpu);
7924 7925 7926
		return;
	}

7927
	capacity = 0;
7928
	min_capacity = ULONG_MAX;
7929

P
Peter Zijlstra 已提交
7930 7931 7932 7933 7934 7935
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

7936
		for_each_cpu(cpu, sched_group_span(sdg)) {
7937
			struct sched_group_capacity *sgc;
7938
			struct rq *rq = cpu_rq(cpu);
7939

7940
			/*
7941
			 * build_sched_domains() -> init_sched_groups_capacity()
7942 7943 7944
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
7945 7946
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
7947
			 *
7948
			 * This avoids capacity from being 0 and
7949 7950 7951
			 * causing divide-by-zero issues on boot.
			 */
			if (unlikely(!rq->sd)) {
7952
				capacity += capacity_of(cpu);
7953 7954 7955
			} else {
				sgc = rq->sd->groups->sgc;
				capacity += sgc->capacity;
7956
			}
7957

7958
			min_capacity = min(capacity, min_capacity);
7959
		}
P
Peter Zijlstra 已提交
7960 7961 7962 7963
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
7964
		 */
P
Peter Zijlstra 已提交
7965 7966 7967

		group = child->groups;
		do {
7968 7969 7970 7971
			struct sched_group_capacity *sgc = group->sgc;

			capacity += sgc->capacity;
			min_capacity = min(sgc->min_capacity, min_capacity);
P
Peter Zijlstra 已提交
7972 7973 7974
			group = group->next;
		} while (group != child->groups);
	}
7975

7976
	sdg->sgc->capacity = capacity;
7977
	sdg->sgc->min_capacity = min_capacity;
7978 7979
}

7980
/*
7981 7982 7983
 * Check whether the capacity of the rq has been noticeably reduced by side
 * activity. The imbalance_pct is used for the threshold.
 * Return true is the capacity is reduced
7984 7985
 */
static inline int
7986
check_cpu_capacity(struct rq *rq, struct sched_domain *sd)
7987
{
7988 7989
	return ((rq->cpu_capacity * sd->imbalance_pct) <
				(rq->cpu_capacity_orig * 100));
7990 7991
}

7992 7993
/*
 * Group imbalance indicates (and tries to solve) the problem where balancing
7994
 * groups is inadequate due to ->cpus_allowed constraints.
7995
 *
7996 7997
 * Imagine a situation of two groups of 4 CPUs each and 4 tasks each with a
 * cpumask covering 1 CPU of the first group and 3 CPUs of the second group.
7998 7999
 * Something like:
 *
8000 8001
 *	{ 0 1 2 3 } { 4 5 6 7 }
 *	        *     * * *
8002 8003 8004
 *
 * If we were to balance group-wise we'd place two tasks in the first group and
 * two tasks in the second group. Clearly this is undesired as it will overload
8005
 * cpu 3 and leave one of the CPUs in the second group unused.
8006 8007
 *
 * The current solution to this issue is detecting the skew in the first group
8008 8009
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
8010 8011
 *
 * When this is so detected; this group becomes a candidate for busiest; see
8012
 * update_sd_pick_busiest(). And calculate_imbalance() and
8013
 * find_busiest_group() avoid some of the usual balance conditions to allow it
8014 8015 8016 8017 8018 8019 8020
 * to create an effective group imbalance.
 *
 * This is a somewhat tricky proposition since the next run might not find the
 * group imbalance and decide the groups need to be balanced again. A most
 * subtle and fragile situation.
 */

8021
static inline int sg_imbalanced(struct sched_group *group)
8022
{
8023
	return group->sgc->imbalance;
8024 8025
}

8026
/*
8027 8028 8029
 * group_has_capacity returns true if the group has spare capacity that could
 * be used by some tasks.
 * We consider that a group has spare capacity if the  * number of task is
8030 8031
 * smaller than the number of CPUs or if the utilization is lower than the
 * available capacity for CFS tasks.
8032 8033 8034 8035 8036
 * For the latter, we use a threshold to stabilize the state, to take into
 * account the variance of the tasks' load and to return true if the available
 * capacity in meaningful for the load balancer.
 * As an example, an available capacity of 1% can appear but it doesn't make
 * any benefit for the load balance.
8037
 */
8038 8039
static inline bool
group_has_capacity(struct lb_env *env, struct sg_lb_stats *sgs)
8040
{
8041 8042
	if (sgs->sum_nr_running < sgs->group_weight)
		return true;
8043

8044
	if ((sgs->group_capacity * 100) >
8045
			(sgs->group_util * env->sd->imbalance_pct))
8046
		return true;
8047

8048 8049 8050 8051 8052 8053 8054 8055 8056 8057 8058 8059 8060 8061 8062 8063
	return false;
}

/*
 *  group_is_overloaded returns true if the group has more tasks than it can
 *  handle.
 *  group_is_overloaded is not equals to !group_has_capacity because a group
 *  with the exact right number of tasks, has no more spare capacity but is not
 *  overloaded so both group_has_capacity and group_is_overloaded return
 *  false.
 */
static inline bool
group_is_overloaded(struct lb_env *env, struct sg_lb_stats *sgs)
{
	if (sgs->sum_nr_running <= sgs->group_weight)
		return false;
8064

8065
	if ((sgs->group_capacity * 100) <
8066
			(sgs->group_util * env->sd->imbalance_pct))
8067
		return true;
8068

8069
	return false;
8070 8071
}

8072 8073 8074 8075 8076 8077 8078 8079 8080 8081 8082
/*
 * group_smaller_cpu_capacity: Returns true if sched_group sg has smaller
 * per-CPU capacity than sched_group ref.
 */
static inline bool
group_smaller_cpu_capacity(struct sched_group *sg, struct sched_group *ref)
{
	return sg->sgc->min_capacity * capacity_margin <
						ref->sgc->min_capacity * 1024;
}

8083 8084 8085
static inline enum
group_type group_classify(struct sched_group *group,
			  struct sg_lb_stats *sgs)
8086
{
8087
	if (sgs->group_no_capacity)
8088 8089 8090 8091 8092 8093 8094 8095
		return group_overloaded;

	if (sg_imbalanced(group))
		return group_imbalanced;

	return group_other;
}

8096
static bool update_nohz_stats(struct rq *rq, bool force)
8097 8098 8099 8100
{
#ifdef CONFIG_NO_HZ_COMMON
	unsigned int cpu = rq->cpu;

8101 8102 8103
	if (!rq->has_blocked_load)
		return false;

8104
	if (!cpumask_test_cpu(cpu, nohz.idle_cpus_mask))
8105
		return false;
8106

8107
	if (!force && !time_after(jiffies, rq->last_blocked_load_update_tick))
8108
		return true;
8109 8110

	update_blocked_averages(cpu);
8111 8112 8113 8114

	return rq->has_blocked_load;
#else
	return false;
8115 8116 8117
#endif
}

8118 8119
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
8120
 * @env: The load balancing environment.
8121 8122 8123 8124
 * @group: sched_group whose statistics are to be updated.
 * @load_idx: Load index of sched_domain of this_cpu for load calc.
 * @local_group: Does group contain this_cpu.
 * @sgs: variable to hold the statistics for this group.
8125
 * @overload: Indicate more than one runnable task for any CPU.
8126
 */
8127 8128
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
8129 8130
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
8131
{
8132
	unsigned long load;
8133
	int i, nr_running;
8134

8135 8136
	memset(sgs, 0, sizeof(*sgs));

8137
	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
8138 8139
		struct rq *rq = cpu_rq(i);

8140
		if ((env->flags & LBF_NOHZ_STATS) && update_nohz_stats(rq, false))
8141
			env->flags |= LBF_NOHZ_AGAIN;
8142

8143
		/* Bias balancing toward CPUs of our domain: */
8144
		if (local_group)
8145
			load = target_load(i, load_idx);
8146
		else
8147 8148 8149
			load = source_load(i, load_idx);

		sgs->group_load += load;
8150
		sgs->group_util += cpu_util(i);
8151
		sgs->sum_nr_running += rq->cfs.h_nr_running;
8152

8153 8154
		nr_running = rq->nr_running;
		if (nr_running > 1)
8155 8156
			*overload = true;

8157 8158 8159 8160
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
8161
		sgs->sum_weighted_load += weighted_cpuload(rq);
8162 8163 8164 8165
		/*
		 * No need to call idle_cpu() if nr_running is not 0
		 */
		if (!nr_running && idle_cpu(i))
8166
			sgs->idle_cpus++;
8167 8168
	}

8169 8170
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
8171
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
8172

8173
	if (sgs->sum_nr_running)
8174
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
8175

8176
	sgs->group_weight = group->group_weight;
8177

8178
	sgs->group_no_capacity = group_is_overloaded(env, sgs);
8179
	sgs->group_type = group_classify(group, sgs);
8180 8181
}

8182 8183
/**
 * update_sd_pick_busiest - return 1 on busiest group
8184
 * @env: The load balancing environment.
8185 8186
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
8187
 * @sgs: sched_group statistics
8188 8189 8190
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
8191 8192 8193
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
8194
 */
8195
static bool update_sd_pick_busiest(struct lb_env *env,
8196 8197
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
8198
				   struct sg_lb_stats *sgs)
8199
{
8200
	struct sg_lb_stats *busiest = &sds->busiest_stat;
8201

8202
	if (sgs->group_type > busiest->group_type)
8203 8204
		return true;

8205 8206 8207 8208 8209 8210
	if (sgs->group_type < busiest->group_type)
		return false;

	if (sgs->avg_load <= busiest->avg_load)
		return false;

8211 8212 8213 8214 8215 8216 8217 8218 8219 8220 8221 8222 8223 8224
	if (!(env->sd->flags & SD_ASYM_CPUCAPACITY))
		goto asym_packing;

	/*
	 * Candidate sg has no more than one task per CPU and
	 * has higher per-CPU capacity. Migrating tasks to less
	 * capable CPUs may harm throughput. Maximize throughput,
	 * power/energy consequences are not considered.
	 */
	if (sgs->sum_nr_running <= sgs->group_weight &&
	    group_smaller_cpu_capacity(sds->local, sg))
		return false;

asym_packing:
8225 8226
	/* This is the busiest node in its class. */
	if (!(env->sd->flags & SD_ASYM_PACKING))
8227 8228
		return true;

8229
	/* No ASYM_PACKING if target CPU is already busy */
8230 8231
	if (env->idle == CPU_NOT_IDLE)
		return true;
8232
	/*
T
Tim Chen 已提交
8233 8234 8235
	 * ASYM_PACKING needs to move all the work to the highest
	 * prority CPUs in the group, therefore mark all groups
	 * of lower priority than ourself as busy.
8236
	 */
T
Tim Chen 已提交
8237 8238
	if (sgs->sum_nr_running &&
	    sched_asym_prefer(env->dst_cpu, sg->asym_prefer_cpu)) {
8239 8240 8241
		if (!sds->busiest)
			return true;

8242
		/* Prefer to move from lowest priority CPU's work */
T
Tim Chen 已提交
8243 8244
		if (sched_asym_prefer(sds->busiest->asym_prefer_cpu,
				      sg->asym_prefer_cpu))
8245 8246 8247 8248 8249 8250
			return true;
	}

	return false;
}

8251 8252 8253 8254 8255 8256 8257 8258 8259 8260 8261 8262 8263 8264 8265 8266 8267 8268 8269 8270 8271 8272 8273 8274 8275 8276 8277 8278 8279 8280
#ifdef CONFIG_NUMA_BALANCING
static inline enum fbq_type fbq_classify_group(struct sg_lb_stats *sgs)
{
	if (sgs->sum_nr_running > sgs->nr_numa_running)
		return regular;
	if (sgs->sum_nr_running > sgs->nr_preferred_running)
		return remote;
	return all;
}

static inline enum fbq_type fbq_classify_rq(struct rq *rq)
{
	if (rq->nr_running > rq->nr_numa_running)
		return regular;
	if (rq->nr_running > rq->nr_preferred_running)
		return remote;
	return all;
}
#else
static inline enum fbq_type fbq_classify_group(struct sg_lb_stats *sgs)
{
	return all;
}

static inline enum fbq_type fbq_classify_rq(struct rq *rq)
{
	return regular;
}
#endif /* CONFIG_NUMA_BALANCING */

8281
/**
8282
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
8283
 * @env: The load balancing environment.
8284 8285
 * @sds: variable to hold the statistics for this sched_domain.
 */
8286
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
8287
{
8288 8289
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
8290
	struct sg_lb_stats *local = &sds->local_stat;
J
Joonsoo Kim 已提交
8291
	struct sg_lb_stats tmp_sgs;
8292
	int load_idx, prefer_sibling = 0;
8293
	bool overload = false;
8294 8295 8296 8297

	if (child && child->flags & SD_PREFER_SIBLING)
		prefer_sibling = 1;

8298
#ifdef CONFIG_NO_HZ_COMMON
8299
	if (env->idle == CPU_NEWLY_IDLE && READ_ONCE(nohz.has_blocked))
8300 8301 8302
		env->flags |= LBF_NOHZ_STATS;
#endif

8303
	load_idx = get_sd_load_idx(env->sd, env->idle);
8304 8305

	do {
J
Joonsoo Kim 已提交
8306
		struct sg_lb_stats *sgs = &tmp_sgs;
8307 8308
		int local_group;

8309
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_span(sg));
J
Joonsoo Kim 已提交
8310 8311
		if (local_group) {
			sds->local = sg;
8312
			sgs = local;
8313 8314

			if (env->idle != CPU_NEWLY_IDLE ||
8315 8316
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
8317
		}
8318

8319 8320
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
8321

8322 8323 8324
		if (local_group)
			goto next_group;

8325 8326
		/*
		 * In case the child domain prefers tasks go to siblings
8327
		 * first, lower the sg capacity so that we'll try
8328 8329
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
8330 8331 8332 8333
		 * these excess tasks. The extra check prevents the case where
		 * you always pull from the heaviest group when it is already
		 * under-utilized (possible with a large weight task outweighs
		 * the tasks on the system).
8334
		 */
8335
		if (prefer_sibling && sds->local &&
8336 8337
		    group_has_capacity(env, local) &&
		    (sgs->sum_nr_running > local->sum_nr_running + 1)) {
8338
			sgs->group_no_capacity = 1;
8339
			sgs->group_type = group_classify(sg, sgs);
8340
		}
8341

8342
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
8343
			sds->busiest = sg;
J
Joonsoo Kim 已提交
8344
			sds->busiest_stat = *sgs;
8345 8346
		}

8347 8348
next_group:
		/* Now, start updating sd_lb_stats */
8349
		sds->total_running += sgs->sum_nr_running;
8350
		sds->total_load += sgs->group_load;
8351
		sds->total_capacity += sgs->group_capacity;
8352

8353
		sg = sg->next;
8354
	} while (sg != env->sd->groups);
8355

8356 8357 8358 8359 8360 8361 8362 8363 8364
#ifdef CONFIG_NO_HZ_COMMON
	if ((env->flags & LBF_NOHZ_AGAIN) &&
	    cpumask_subset(nohz.idle_cpus_mask, sched_domain_span(env->sd))) {

		WRITE_ONCE(nohz.next_blocked,
			   jiffies + msecs_to_jiffies(LOAD_AVG_PERIOD));
	}
#endif

8365 8366
	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
8367 8368 8369 8370 8371 8372

	if (!env->sd->parent) {
		/* update overload indicator if we are at root domain */
		if (env->dst_rq->rd->overload != overload)
			env->dst_rq->rd->overload = overload;
	}
8373 8374 8375 8376
}

/**
 * check_asym_packing - Check to see if the group is packed into the
8377
 *			sched domain.
8378 8379 8380 8381 8382 8383 8384 8385 8386 8387 8388 8389 8390 8391
 *
 * This is primarily intended to used at the sibling level.  Some
 * cores like POWER7 prefer to use lower numbered SMT threads.  In the
 * case of POWER7, it can move to lower SMT modes only when higher
 * threads are idle.  When in lower SMT modes, the threads will
 * perform better since they share less core resources.  Hence when we
 * have idle threads, we want them to be the higher ones.
 *
 * This packing function is run on idle threads.  It checks to see if
 * the busiest CPU in this domain (core in the P7 case) has a higher
 * CPU number than the packing function is being run on.  Here we are
 * assuming lower CPU number will be equivalent to lower a SMT thread
 * number.
 *
8392
 * Return: 1 when packing is required and a task should be moved to
8393
 * this CPU.  The amount of the imbalance is returned in env->imbalance.
8394
 *
8395
 * @env: The load balancing environment.
8396 8397
 * @sds: Statistics of the sched_domain which is to be packed
 */
8398
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
8399 8400 8401
{
	int busiest_cpu;

8402
	if (!(env->sd->flags & SD_ASYM_PACKING))
8403 8404
		return 0;

8405 8406 8407
	if (env->idle == CPU_NOT_IDLE)
		return 0;

8408 8409 8410
	if (!sds->busiest)
		return 0;

T
Tim Chen 已提交
8411 8412
	busiest_cpu = sds->busiest->asym_prefer_cpu;
	if (sched_asym_prefer(busiest_cpu, env->dst_cpu))
8413 8414
		return 0;

8415
	env->imbalance = DIV_ROUND_CLOSEST(
8416
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
8417
		SCHED_CAPACITY_SCALE);
8418

8419
	return 1;
8420 8421 8422 8423 8424 8425
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
8426
 * @env: The load balancing environment.
8427 8428
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
8429 8430
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
8431
{
8432
	unsigned long tmp, capa_now = 0, capa_move = 0;
8433
	unsigned int imbn = 2;
8434
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
8435
	struct sg_lb_stats *local, *busiest;
8436

J
Joonsoo Kim 已提交
8437 8438
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
8439

J
Joonsoo Kim 已提交
8440 8441 8442 8443
	if (!local->sum_nr_running)
		local->load_per_task = cpu_avg_load_per_task(env->dst_cpu);
	else if (busiest->load_per_task > local->load_per_task)
		imbn = 1;
8444

J
Joonsoo Kim 已提交
8445
	scaled_busy_load_per_task =
8446
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
8447
		busiest->group_capacity;
J
Joonsoo Kim 已提交
8448

8449 8450
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
8451
		env->imbalance = busiest->load_per_task;
8452 8453 8454 8455 8456
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
8457
	 * however we may be able to increase total CPU capacity used by
8458 8459 8460
	 * moving them.
	 */

8461
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
8462
			min(busiest->load_per_task, busiest->avg_load);
8463
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
8464
			min(local->load_per_task, local->avg_load);
8465
	capa_now /= SCHED_CAPACITY_SCALE;
8466 8467

	/* Amount of load we'd subtract */
8468
	if (busiest->avg_load > scaled_busy_load_per_task) {
8469
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
8470
			    min(busiest->load_per_task,
8471
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
8472
	}
8473 8474

	/* Amount of load we'd add */
8475
	if (busiest->avg_load * busiest->group_capacity <
8476
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
8477 8478
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
8479
	} else {
8480
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
8481
		      local->group_capacity;
J
Joonsoo Kim 已提交
8482
	}
8483
	capa_move += local->group_capacity *
8484
		    min(local->load_per_task, local->avg_load + tmp);
8485
	capa_move /= SCHED_CAPACITY_SCALE;
8486 8487

	/* Move if we gain throughput */
8488
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
8489
		env->imbalance = busiest->load_per_task;
8490 8491 8492 8493 8494
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
8495
 * @env: load balance environment
8496 8497
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
8498
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
8499
{
8500
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
8501 8502 8503 8504
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
8505

8506
	if (busiest->group_type == group_imbalanced) {
8507 8508
		/*
		 * In the group_imb case we cannot rely on group-wide averages
8509
		 * to ensure CPU-load equilibrium, look at wider averages. XXX
8510
		 */
J
Joonsoo Kim 已提交
8511 8512
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
8513 8514
	}

8515
	/*
8516 8517 8518 8519
	 * Avg load of busiest sg can be less and avg load of local sg can
	 * be greater than avg load across all sgs of sd because avg load
	 * factors in sg capacity and sgs with smaller group_type are
	 * skipped when updating the busiest sg:
8520
	 */
8521 8522
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
8523 8524
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
8525 8526
	}

8527
	/*
8528
	 * If there aren't any idle CPUs, avoid creating some.
8529 8530 8531
	 */
	if (busiest->group_type == group_overloaded &&
	    local->group_type   == group_overloaded) {
8532
		load_above_capacity = busiest->sum_nr_running * SCHED_CAPACITY_SCALE;
8533
		if (load_above_capacity > busiest->group_capacity) {
8534
			load_above_capacity -= busiest->group_capacity;
8535
			load_above_capacity *= scale_load_down(NICE_0_LOAD);
8536 8537
			load_above_capacity /= busiest->group_capacity;
		} else
8538
			load_above_capacity = ~0UL;
8539 8540 8541
	}

	/*
8542
	 * We're trying to get all the CPUs to the average_load, so we don't
8543
	 * want to push ourselves above the average load, nor do we wish to
8544
	 * reduce the max loaded CPU below the average load. At the same time,
8545 8546
	 * we also don't want to reduce the group load below the group
	 * capacity. Thus we look for the minimum possible imbalance.
8547
	 */
8548
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
8549 8550

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
8551
	env->imbalance = min(
8552 8553
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
8554
	) / SCHED_CAPACITY_SCALE;
8555 8556 8557

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
8558
	 * there is no guarantee that any tasks will be moved so we'll have
8559 8560 8561
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
8562
	if (env->imbalance < busiest->load_per_task)
8563
		return fix_small_imbalance(env, sds);
8564
}
8565

8566 8567 8568 8569
/******* find_busiest_group() helpers end here *********************/

/**
 * find_busiest_group - Returns the busiest group within the sched_domain
8570
 * if there is an imbalance.
8571 8572 8573 8574
 *
 * Also calculates the amount of weighted load which should be moved
 * to restore balance.
 *
8575
 * @env: The load balancing environment.
8576
 *
8577
 * Return:	- The busiest group if imbalance exists.
8578
 */
J
Joonsoo Kim 已提交
8579
static struct sched_group *find_busiest_group(struct lb_env *env)
8580
{
J
Joonsoo Kim 已提交
8581
	struct sg_lb_stats *local, *busiest;
8582 8583
	struct sd_lb_stats sds;

8584
	init_sd_lb_stats(&sds);
8585 8586 8587 8588 8589

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
8590
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
8591 8592
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
8593

8594
	/* ASYM feature bypasses nice load balance check */
8595
	if (check_asym_packing(env, &sds))
8596 8597
		return sds.busiest;

8598
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
8599
	if (!sds.busiest || busiest->sum_nr_running == 0)
8600 8601
		goto out_balanced;

8602
	/* XXX broken for overlapping NUMA groups */
8603 8604
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
8605

P
Peter Zijlstra 已提交
8606 8607
	/*
	 * If the busiest group is imbalanced the below checks don't
8608
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
8609 8610
	 * isn't true due to cpus_allowed constraints and the like.
	 */
8611
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
8612 8613
		goto force_balance;

8614 8615 8616 8617 8618
	/*
	 * When dst_cpu is idle, prevent SMP nice and/or asymmetric group
	 * capacities from resulting in underutilization due to avg_load.
	 */
	if (env->idle != CPU_NOT_IDLE && group_has_capacity(env, local) &&
8619
	    busiest->group_no_capacity)
8620 8621
		goto force_balance;

8622
	/*
8623
	 * If the local group is busier than the selected busiest group
8624 8625
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
8626
	if (local->avg_load >= busiest->avg_load)
8627 8628
		goto out_balanced;

8629 8630 8631 8632
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
8633
	if (local->avg_load >= sds.avg_load)
8634 8635
		goto out_balanced;

8636
	if (env->idle == CPU_IDLE) {
8637
		/*
8638
		 * This CPU is idle. If the busiest group is not overloaded
8639
		 * and there is no imbalance between this and busiest group
8640
		 * wrt idle CPUs, it is balanced. The imbalance becomes
8641 8642
		 * significant if the diff is greater than 1 otherwise we
		 * might end up to just move the imbalance on another group
8643
		 */
8644 8645
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
8646
			goto out_balanced;
8647 8648 8649 8650 8651
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
8652 8653
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
8654
			goto out_balanced;
8655
	}
8656

8657
force_balance:
8658
	/* Looks like there is an imbalance. Compute it */
8659
	calculate_imbalance(env, &sds);
8660 8661 8662
	return sds.busiest;

out_balanced:
8663
	env->imbalance = 0;
8664 8665 8666 8667
	return NULL;
}

/*
8668
 * find_busiest_queue - find the busiest runqueue among the CPUs in the group.
8669
 */
8670
static struct rq *find_busiest_queue(struct lb_env *env,
8671
				     struct sched_group *group)
8672 8673
{
	struct rq *busiest = NULL, *rq;
8674
	unsigned long busiest_load = 0, busiest_capacity = 1;
8675 8676
	int i;

8677
	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
8678
		unsigned long capacity, wl;
8679 8680 8681 8682
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
8683

8684 8685 8686 8687 8688 8689 8690 8691 8692 8693 8694 8695 8696 8697 8698 8699 8700 8701 8702 8703 8704 8705
		/*
		 * We classify groups/runqueues into three groups:
		 *  - regular: there are !numa tasks
		 *  - remote:  there are numa tasks that run on the 'wrong' node
		 *  - all:     there is no distinction
		 *
		 * In order to avoid migrating ideally placed numa tasks,
		 * ignore those when there's better options.
		 *
		 * If we ignore the actual busiest queue to migrate another
		 * task, the next balance pass can still reduce the busiest
		 * queue by moving tasks around inside the node.
		 *
		 * If we cannot move enough load due to this classification
		 * the next pass will adjust the group classification and
		 * allow migration of more tasks.
		 *
		 * Both cases only affect the total convergence complexity.
		 */
		if (rt > env->fbq_type)
			continue;

8706
		capacity = capacity_of(i);
8707

8708
		wl = weighted_cpuload(rq);
8709

8710 8711
		/*
		 * When comparing with imbalance, use weighted_cpuload()
8712
		 * which is not scaled with the CPU capacity.
8713
		 */
8714 8715 8716

		if (rq->nr_running == 1 && wl > env->imbalance &&
		    !check_cpu_capacity(rq, env->sd))
8717 8718
			continue;

8719
		/*
8720 8721 8722
		 * For the load comparisons with the other CPU's, consider
		 * the weighted_cpuload() scaled with the CPU capacity, so
		 * that the load can be moved away from the CPU that is
8723
		 * potentially running at a lower capacity.
8724
		 *
8725
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
8726
		 * multiplication to rid ourselves of the division works out
8727 8728
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
8729
		 */
8730
		if (wl * busiest_capacity > busiest_load * capacity) {
8731
			busiest_load = wl;
8732
			busiest_capacity = capacity;
8733 8734 8735 8736 8737 8738 8739 8740 8741 8742 8743 8744 8745
			busiest = rq;
		}
	}

	return busiest;
}

/*
 * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but
 * so long as it is large enough.
 */
#define MAX_PINNED_INTERVAL	512

8746
static int need_active_balance(struct lb_env *env)
8747
{
8748 8749 8750
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
8751 8752 8753

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
T
Tim Chen 已提交
8754 8755
		 * lower priority CPUs in order to pack all tasks in the
		 * highest priority CPUs.
8756
		 */
T
Tim Chen 已提交
8757 8758
		if ((sd->flags & SD_ASYM_PACKING) &&
		    sched_asym_prefer(env->dst_cpu, env->src_cpu))
8759
			return 1;
8760 8761
	}

8762 8763 8764 8765 8766 8767 8768 8769 8770 8771 8772 8773 8774
	/*
	 * The dst_cpu is idle and the src_cpu CPU has only 1 CFS task.
	 * It's worth migrating the task if the src_cpu's capacity is reduced
	 * because of other sched_class or IRQs if more capacity stays
	 * available on dst_cpu.
	 */
	if ((env->idle != CPU_NOT_IDLE) &&
	    (env->src_rq->cfs.h_nr_running == 1)) {
		if ((check_cpu_capacity(env->src_rq, sd)) &&
		    (capacity_of(env->src_cpu)*sd->imbalance_pct < capacity_of(env->dst_cpu)*100))
			return 1;
	}

8775 8776 8777
	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

8778 8779
static int active_load_balance_cpu_stop(void *data);

8780 8781 8782 8783 8784
static int should_we_balance(struct lb_env *env)
{
	struct sched_group *sg = env->sd->groups;
	int cpu, balance_cpu = -1;

8785 8786 8787 8788 8789 8790 8791
	/*
	 * Ensure the balancing environment is consistent; can happen
	 * when the softirq triggers 'during' hotplug.
	 */
	if (!cpumask_test_cpu(env->dst_cpu, env->cpus))
		return 0;

8792
	/*
8793
	 * In the newly idle case, we will allow all the CPUs
8794 8795 8796 8797 8798
	 * to do the newly idle load balance.
	 */
	if (env->idle == CPU_NEWLY_IDLE)
		return 1;

8799
	/* Try to find first idle CPU */
8800
	for_each_cpu_and(cpu, group_balance_mask(sg), env->cpus) {
8801
		if (!idle_cpu(cpu))
8802 8803 8804 8805 8806 8807 8808 8809 8810 8811
			continue;

		balance_cpu = cpu;
		break;
	}

	if (balance_cpu == -1)
		balance_cpu = group_balance_cpu(sg);

	/*
8812
	 * First idle CPU or the first CPU(busiest) in this sched group
8813 8814
	 * is eligible for doing load balancing at this and above domains.
	 */
8815
	return balance_cpu == env->dst_cpu;
8816 8817
}

8818 8819 8820 8821 8822 8823
/*
 * Check this_cpu to ensure it is balanced within domain. Attempt to move
 * tasks if there is an imbalance.
 */
static int load_balance(int this_cpu, struct rq *this_rq,
			struct sched_domain *sd, enum cpu_idle_type idle,
8824
			int *continue_balancing)
8825
{
8826
	int ld_moved, cur_ld_moved, active_balance = 0;
8827
	struct sched_domain *sd_parent = sd->parent;
8828 8829
	struct sched_group *group;
	struct rq *busiest;
8830
	struct rq_flags rf;
8831
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
8832

8833 8834
	struct lb_env env = {
		.sd		= sd,
8835 8836
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
8837
		.dst_grpmask    = sched_group_span(sd->groups),
8838
		.idle		= idle,
8839
		.loop_break	= sched_nr_migrate_break,
8840
		.cpus		= cpus,
8841
		.fbq_type	= all,
8842
		.tasks		= LIST_HEAD_INIT(env.tasks),
8843 8844
	};

8845
	cpumask_and(cpus, sched_domain_span(sd), cpu_active_mask);
8846

8847
	schedstat_inc(sd->lb_count[idle]);
8848 8849

redo:
8850 8851
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
8852
		goto out_balanced;
8853
	}
8854

8855
	group = find_busiest_group(&env);
8856
	if (!group) {
8857
		schedstat_inc(sd->lb_nobusyg[idle]);
8858 8859 8860
		goto out_balanced;
	}

8861
	busiest = find_busiest_queue(&env, group);
8862
	if (!busiest) {
8863
		schedstat_inc(sd->lb_nobusyq[idle]);
8864 8865 8866
		goto out_balanced;
	}

8867
	BUG_ON(busiest == env.dst_rq);
8868

8869
	schedstat_add(sd->lb_imbalance[idle], env.imbalance);
8870

8871 8872 8873
	env.src_cpu = busiest->cpu;
	env.src_rq = busiest;

8874 8875 8876 8877 8878 8879 8880 8881
	ld_moved = 0;
	if (busiest->nr_running > 1) {
		/*
		 * Attempt to move tasks. If find_busiest_group has found
		 * an imbalance but busiest->nr_running <= 1, the group is
		 * still unbalanced. ld_moved simply stays zero, so it is
		 * correctly treated as an imbalance.
		 */
8882
		env.flags |= LBF_ALL_PINNED;
8883
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
8884

8885
more_balance:
8886
		rq_lock_irqsave(busiest, &rf);
8887
		update_rq_clock(busiest);
8888 8889 8890 8891 8892

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
8893
		cur_ld_moved = detach_tasks(&env);
8894 8895

		/*
8896 8897 8898 8899 8900
		 * We've detached some tasks from busiest_rq. Every
		 * task is masked "TASK_ON_RQ_MIGRATING", so we can safely
		 * unlock busiest->lock, and we are able to be sure
		 * that nobody can manipulate the tasks in parallel.
		 * See task_rq_lock() family for the details.
8901
		 */
8902

8903
		rq_unlock(busiest, &rf);
8904 8905 8906 8907 8908 8909

		if (cur_ld_moved) {
			attach_tasks(&env);
			ld_moved += cur_ld_moved;
		}

8910
		local_irq_restore(rf.flags);
8911

8912 8913 8914 8915 8916
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

8917 8918 8919 8920
		/*
		 * Revisit (affine) tasks on src_cpu that couldn't be moved to
		 * us and move them to an alternate dst_cpu in our sched_group
		 * where they can run. The upper limit on how many times we
8921
		 * iterate on same src_cpu is dependent on number of CPUs in our
8922 8923 8924 8925 8926 8927 8928 8929 8930 8931 8932 8933 8934 8935
		 * sched_group.
		 *
		 * This changes load balance semantics a bit on who can move
		 * load to a given_cpu. In addition to the given_cpu itself
		 * (or a ilb_cpu acting on its behalf where given_cpu is
		 * nohz-idle), we now have balance_cpu in a position to move
		 * load to given_cpu. In rare situations, this may cause
		 * conflicts (balance_cpu and given_cpu/ilb_cpu deciding
		 * _independently_ and at _same_ time to move some load to
		 * given_cpu) causing exceess load to be moved to given_cpu.
		 * This however should not happen so much in practice and
		 * moreover subsequent load balance cycles should correct the
		 * excess load moved.
		 */
8936
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
8937

8938
			/* Prevent to re-select dst_cpu via env's CPUs */
8939 8940
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

8941
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
8942
			env.dst_cpu	 = env.new_dst_cpu;
8943
			env.flags	&= ~LBF_DST_PINNED;
8944 8945
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
8946

8947 8948 8949 8950 8951 8952
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
8953

8954 8955 8956 8957
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
8958
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
8959

8960
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
8961 8962 8963
				*group_imbalance = 1;
		}

8964
		/* All tasks on this runqueue were pinned by CPU affinity */
8965
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
8966
			cpumask_clear_cpu(cpu_of(busiest), cpus);
8967 8968 8969 8970 8971 8972 8973 8974 8975
			/*
			 * Attempting to continue load balancing at the current
			 * sched_domain level only makes sense if there are
			 * active CPUs remaining as possible busiest CPUs to
			 * pull load from which are not contained within the
			 * destination group that is receiving any migrated
			 * load.
			 */
			if (!cpumask_subset(cpus, env.dst_grpmask)) {
8976 8977
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
8978
				goto redo;
8979
			}
8980
			goto out_all_pinned;
8981 8982 8983 8984
		}
	}

	if (!ld_moved) {
8985
		schedstat_inc(sd->lb_failed[idle]);
8986 8987 8988 8989 8990 8991 8992 8993
		/*
		 * Increment the failure counter only on periodic balance.
		 * We do not want newidle balance, which can be very
		 * frequent, pollute the failure counter causing
		 * excessive cache_hot migrations and active balances.
		 */
		if (idle != CPU_NEWLY_IDLE)
			sd->nr_balance_failed++;
8994

8995
		if (need_active_balance(&env)) {
8996 8997
			unsigned long flags;

8998 8999
			raw_spin_lock_irqsave(&busiest->lock, flags);

9000 9001 9002 9003
			/*
			 * Don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest CPU can't be
			 * moved to this_cpu:
9004
			 */
9005
			if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) {
9006 9007
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
9008
				env.flags |= LBF_ALL_PINNED;
9009 9010 9011
				goto out_one_pinned;
			}

9012 9013 9014 9015 9016
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
9017 9018 9019 9020 9021 9022
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
9023

9024
			if (active_balance) {
9025 9026 9027
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
9028
			}
9029

9030
			/* We've kicked active balancing, force task migration. */
9031 9032 9033 9034 9035 9036 9037 9038 9039 9040 9041 9042 9043
			sd->nr_balance_failed = sd->cache_nice_tries+1;
		}
	} else
		sd->nr_balance_failed = 0;

	if (likely(!active_balance)) {
		/* We were unbalanced, so reset the balancing interval */
		sd->balance_interval = sd->min_interval;
	} else {
		/*
		 * If we've begun active balancing, start to back off. This
		 * case may not be covered by the all_pinned logic if there
		 * is only 1 task on the busy runqueue (because we don't call
9044
		 * detach_tasks).
9045 9046 9047 9048 9049 9050 9051 9052
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
9053 9054 9055 9056 9057 9058 9059 9060 9061 9062 9063 9064 9065 9066 9067 9068 9069
	/*
	 * We reach balance although we may have faced some affinity
	 * constraints. Clear the imbalance flag if it was set.
	 */
	if (sd_parent) {
		int *group_imbalance = &sd_parent->groups->sgc->imbalance;

		if (*group_imbalance)
			*group_imbalance = 0;
	}

out_all_pinned:
	/*
	 * We reach balance because all tasks are pinned at this level so
	 * we can't migrate them. Let the imbalance flag set so parent level
	 * can try to migrate them.
	 */
9070
	schedstat_inc(sd->lb_balanced[idle]);
9071 9072 9073 9074 9075

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
9076
	if (((env.flags & LBF_ALL_PINNED) &&
9077
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
9078 9079 9080
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

9081
	ld_moved = 0;
9082 9083 9084 9085
out:
	return ld_moved;
}

9086 9087 9088 9089 9090 9091 9092 9093 9094 9095 9096 9097 9098 9099 9100 9101
static inline unsigned long
get_sd_balance_interval(struct sched_domain *sd, int cpu_busy)
{
	unsigned long interval = sd->balance_interval;

	if (cpu_busy)
		interval *= sd->busy_factor;

	/* scale ms to jiffies */
	interval = msecs_to_jiffies(interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);

	return interval;
}

static inline void
9102
update_next_balance(struct sched_domain *sd, unsigned long *next_balance)
9103 9104 9105
{
	unsigned long interval, next;

9106 9107
	/* used by idle balance, so cpu_busy = 0 */
	interval = get_sd_balance_interval(sd, 0);
9108 9109 9110 9111 9112 9113
	next = sd->last_balance + interval;

	if (time_after(*next_balance, next))
		*next_balance = next;
}

9114
/*
9115
 * active_load_balance_cpu_stop is run by the CPU stopper. It pushes
9116 9117 9118
 * running tasks off the busiest CPU onto idle CPUs. It requires at
 * least 1 task to be running on each physical CPU where possible, and
 * avoids physical / logical imbalances.
9119
 */
9120
static int active_load_balance_cpu_stop(void *data)
9121
{
9122 9123
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
9124
	int target_cpu = busiest_rq->push_cpu;
9125
	struct rq *target_rq = cpu_rq(target_cpu);
9126
	struct sched_domain *sd;
9127
	struct task_struct *p = NULL;
9128
	struct rq_flags rf;
9129

9130
	rq_lock_irq(busiest_rq, &rf);
9131 9132 9133 9134 9135 9136 9137
	/*
	 * Between queueing the stop-work and running it is a hole in which
	 * CPUs can become inactive. We should not move tasks from or to
	 * inactive CPUs.
	 */
	if (!cpu_active(busiest_cpu) || !cpu_active(target_cpu))
		goto out_unlock;
9138

9139
	/* Make sure the requested CPU hasn't gone down in the meantime: */
9140 9141 9142
	if (unlikely(busiest_cpu != smp_processor_id() ||
		     !busiest_rq->active_balance))
		goto out_unlock;
9143 9144 9145

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
9146
		goto out_unlock;
9147 9148 9149 9150

	/*
	 * This condition is "impossible", if it occurs
	 * we need to fix it. Originally reported by
9151
	 * Bjorn Helgaas on a 128-CPU setup.
9152 9153 9154 9155
	 */
	BUG_ON(busiest_rq == target_rq);

	/* Search for an sd spanning us and the target CPU. */
9156
	rcu_read_lock();
9157 9158 9159 9160 9161 9162 9163
	for_each_domain(target_cpu, sd) {
		if ((sd->flags & SD_LOAD_BALANCE) &&
		    cpumask_test_cpu(busiest_cpu, sched_domain_span(sd)))
				break;
	}

	if (likely(sd)) {
9164 9165
		struct lb_env env = {
			.sd		= sd,
9166 9167 9168 9169
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
9170
			.idle		= CPU_IDLE,
9171 9172 9173 9174 9175 9176 9177
			/*
			 * can_migrate_task() doesn't need to compute new_dst_cpu
			 * for active balancing. Since we have CPU_IDLE, but no
			 * @dst_grpmask we need to make that test go away with lying
			 * about DST_PINNED.
			 */
			.flags		= LBF_DST_PINNED,
9178 9179
		};

9180
		schedstat_inc(sd->alb_count);
9181
		update_rq_clock(busiest_rq);
9182

9183
		p = detach_one_task(&env);
9184
		if (p) {
9185
			schedstat_inc(sd->alb_pushed);
9186 9187 9188
			/* Active balancing done, reset the failure counter. */
			sd->nr_balance_failed = 0;
		} else {
9189
			schedstat_inc(sd->alb_failed);
9190
		}
9191
	}
9192
	rcu_read_unlock();
9193 9194
out_unlock:
	busiest_rq->active_balance = 0;
9195
	rq_unlock(busiest_rq, &rf);
9196 9197 9198 9199 9200 9201

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

9202
	return 0;
9203 9204
}

9205 9206 9207 9208 9209 9210 9211 9212 9213 9214 9215 9216 9217 9218 9219 9220 9221 9222 9223 9224 9225 9226 9227 9228 9229 9230 9231 9232 9233 9234 9235 9236 9237 9238 9239 9240 9241 9242 9243 9244 9245 9246 9247 9248 9249 9250 9251 9252 9253 9254 9255 9256 9257 9258 9259 9260 9261 9262 9263 9264 9265 9266 9267 9268 9269 9270 9271 9272 9273 9274 9275 9276 9277 9278 9279 9280 9281 9282 9283 9284 9285 9286 9287 9288 9289 9290 9291 9292 9293 9294 9295 9296 9297 9298 9299 9300 9301 9302 9303 9304 9305 9306 9307 9308 9309 9310 9311 9312 9313 9314 9315 9316 9317 9318 9319 9320 9321 9322
static DEFINE_SPINLOCK(balancing);

/*
 * Scale the max load_balance interval with the number of CPUs in the system.
 * This trades load-balance latency on larger machines for less cross talk.
 */
void update_max_interval(void)
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
 * Balancing parameters are set up in init_sched_domains.
 */
static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle)
{
	int continue_balancing = 1;
	int cpu = rq->cpu;
	unsigned long interval;
	struct sched_domain *sd;
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;

	rcu_read_lock();
	for_each_domain(cpu, sd) {
		/*
		 * Decay the newidle max times here because this is a regular
		 * visit to all the domains. Decay ~1% per second.
		 */
		if (time_after(jiffies, sd->next_decay_max_lb_cost)) {
			sd->max_newidle_lb_cost =
				(sd->max_newidle_lb_cost * 253) / 256;
			sd->next_decay_max_lb_cost = jiffies + HZ;
			need_decay = 1;
		}
		max_cost += sd->max_newidle_lb_cost;

		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

		/*
		 * Stop the load balance at this level. There is another
		 * CPU in our sched group which is doing load balancing more
		 * actively.
		 */
		if (!continue_balancing) {
			if (need_decay)
				continue;
			break;
		}

		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);

		need_serialize = sd->flags & SD_SERIALIZE;
		if (need_serialize) {
			if (!spin_trylock(&balancing))
				goto out;
		}

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
				/*
				 * The LBF_DST_PINNED logic could have changed
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
				 */
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
			}
			sd->last_balance = jiffies;
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
		}
		if (need_serialize)
			spin_unlock(&balancing);
out:
		if (time_after(next_balance, sd->last_balance + interval)) {
			next_balance = sd->last_balance + interval;
			update_next_balance = 1;
		}
	}
	if (need_decay) {
		/*
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
		 */
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
	}
	rcu_read_unlock();

	/*
	 * next_balance will be updated only when there is a need.
	 * When the cpu is attached to null domain for ex, it will not be
	 * updated.
	 */
	if (likely(update_next_balance)) {
		rq->next_balance = next_balance;

#ifdef CONFIG_NO_HZ_COMMON
		/*
		 * If this CPU has been elected to perform the nohz idle
		 * balance. Other idle CPUs have already rebalanced with
		 * nohz_idle_balance() and nohz.next_balance has been
		 * updated accordingly. This CPU is now running the idle load
		 * balance for itself and we need to update the
		 * nohz.next_balance accordingly.
		 */
		if ((idle == CPU_IDLE) && time_after(nohz.next_balance, rq->next_balance))
			nohz.next_balance = rq->next_balance;
#endif
	}
}

9323 9324 9325 9326 9327
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

9328
#ifdef CONFIG_NO_HZ_COMMON
9329 9330 9331 9332 9333 9334
/*
 * idle load balancing details
 * - When one of the busy CPUs notice that there may be an idle rebalancing
 *   needed, they will kick the idle load balancer, which then does idle
 *   load balancing for all the idle CPUs.
 */
9335

9336
static inline int find_new_ilb(void)
9337
{
9338
	int ilb = cpumask_first(nohz.idle_cpus_mask);
9339

9340 9341 9342 9343
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
9344 9345
}

9346 9347 9348 9349 9350
/*
 * Kick a CPU to do the nohz balancing, if it is time for it. We pick the
 * nohz_load_balancer CPU (if there is one) otherwise fallback to any idle
 * CPU (if there is one).
 */
9351
static void kick_ilb(unsigned int flags)
9352 9353 9354 9355 9356
{
	int ilb_cpu;

	nohz.next_balance++;

9357
	ilb_cpu = find_new_ilb();
9358

9359 9360
	if (ilb_cpu >= nr_cpu_ids)
		return;
9361

9362
	flags = atomic_fetch_or(flags, nohz_flags(ilb_cpu));
P
Peter Zijlstra 已提交
9363
	if (flags & NOHZ_KICK_MASK)
9364
		return;
9365

9366 9367
	/*
	 * Use smp_send_reschedule() instead of resched_cpu().
9368
	 * This way we generate a sched IPI on the target CPU which
9369 9370 9371 9372
	 * is idle. And the softirq performing nohz idle load balance
	 * will be run before returning from the IPI.
	 */
	smp_send_reschedule(ilb_cpu);
9373 9374 9375 9376 9377 9378 9379 9380 9381 9382 9383 9384 9385 9386 9387 9388 9389 9390 9391
}

/*
 * Current heuristic for kicking the idle load balancer in the presence
 * of an idle cpu in the system.
 *   - This rq has more than one task.
 *   - This rq has at least one CFS task and the capacity of the CPU is
 *     significantly reduced because of RT tasks or IRQs.
 *   - At parent of LLC scheduler domain level, this cpu's scheduler group has
 *     multiple busy cpu.
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
 */
static void nohz_balancer_kick(struct rq *rq)
{
	unsigned long now = jiffies;
	struct sched_domain_shared *sds;
	struct sched_domain *sd;
	int nr_busy, i, cpu = rq->cpu;
9392
	unsigned int flags = 0;
9393 9394 9395 9396 9397 9398 9399 9400

	if (unlikely(rq->idle_balance))
		return;

	/*
	 * We may be recently in ticked or tickless idle mode. At the first
	 * busy tick after returning from idle, we will update the busy stats.
	 */
9401
	nohz_balance_exit_idle(rq);
9402 9403 9404 9405 9406 9407 9408 9409

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
		return;

9410 9411
	if (READ_ONCE(nohz.has_blocked) &&
	    time_after(now, READ_ONCE(nohz.next_blocked)))
9412 9413
		flags = NOHZ_STATS_KICK;

9414
	if (time_before(now, nohz.next_balance))
9415
		goto out;
9416 9417

	if (rq->nr_running >= 2) {
9418
		flags = NOHZ_KICK_MASK;
9419 9420 9421 9422 9423 9424 9425 9426 9427 9428 9429 9430
		goto out;
	}

	rcu_read_lock();
	sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));
	if (sds) {
		/*
		 * XXX: write a coherent comment on why we do this.
		 * See also: http://lkml.kernel.org/r/20111202010832.602203411@sbsiddha-desk.sc.intel.com
		 */
		nr_busy = atomic_read(&sds->nr_busy_cpus);
		if (nr_busy > 1) {
9431
			flags = NOHZ_KICK_MASK;
9432 9433 9434 9435 9436 9437 9438 9439 9440
			goto unlock;
		}

	}

	sd = rcu_dereference(rq->sd);
	if (sd) {
		if ((rq->cfs.h_nr_running >= 1) &&
				check_cpu_capacity(rq, sd)) {
9441
			flags = NOHZ_KICK_MASK;
9442 9443 9444 9445 9446 9447 9448 9449 9450 9451 9452 9453
			goto unlock;
		}
	}

	sd = rcu_dereference(per_cpu(sd_asym, cpu));
	if (sd) {
		for_each_cpu(i, sched_domain_span(sd)) {
			if (i == cpu ||
			    !cpumask_test_cpu(i, nohz.idle_cpus_mask))
				continue;

			if (sched_asym_prefer(i, cpu)) {
9454
				flags = NOHZ_KICK_MASK;
9455 9456 9457 9458 9459 9460 9461
				goto unlock;
			}
		}
	}
unlock:
	rcu_read_unlock();
out:
9462 9463
	if (flags)
		kick_ilb(flags);
9464 9465
}

9466
static void set_cpu_sd_state_busy(int cpu)
9467
{
9468
	struct sched_domain *sd;
9469

9470 9471
	rcu_read_lock();
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
9472

9473 9474 9475 9476 9477 9478 9479
	if (!sd || !sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 0;

	atomic_inc(&sd->shared->nr_busy_cpus);
unlock:
	rcu_read_unlock();
9480 9481
}

9482 9483 9484 9485 9486 9487 9488 9489 9490 9491 9492 9493 9494 9495 9496
void nohz_balance_exit_idle(struct rq *rq)
{
	SCHED_WARN_ON(rq != this_rq());

	if (likely(!rq->nohz_tick_stopped))
		return;

	rq->nohz_tick_stopped = 0;
	cpumask_clear_cpu(rq->cpu, nohz.idle_cpus_mask);
	atomic_dec(&nohz.nr_cpus);

	set_cpu_sd_state_busy(rq->cpu);
}

static void set_cpu_sd_state_idle(int cpu)
9497 9498 9499 9500
{
	struct sched_domain *sd;

	rcu_read_lock();
9501
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
V
Vincent Guittot 已提交
9502 9503 9504 9505 9506

	if (!sd || sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 1;

9507
	atomic_dec(&sd->shared->nr_busy_cpus);
V
Vincent Guittot 已提交
9508
unlock:
9509 9510 9511
	rcu_read_unlock();
}

9512
/*
9513
 * This routine will record that the CPU is going idle with tick stopped.
9514
 * This info will be used in performing idle load balancing in the future.
9515
 */
9516
void nohz_balance_enter_idle(int cpu)
9517
{
9518 9519 9520 9521
	struct rq *rq = cpu_rq(cpu);

	SCHED_WARN_ON(cpu != smp_processor_id());

9522
	/* If this CPU is going down, then nothing needs to be done: */
9523 9524 9525
	if (!cpu_active(cpu))
		return;

9526
	/* Spare idle load balancing on CPUs that don't want to be disturbed: */
9527
	if (!housekeeping_cpu(cpu, HK_FLAG_SCHED))
9528 9529
		return;

9530 9531 9532 9533 9534 9535 9536 9537 9538 9539 9540 9541 9542
	/*
	 * Can be set safely without rq->lock held
	 * If a clear happens, it will have evaluated last additions because
	 * rq->lock is held during the check and the clear
	 */
	rq->has_blocked_load = 1;

	/*
	 * The tick is still stopped but load could have been added in the
	 * meantime. We set the nohz.has_blocked flag to trig a check of the
	 * *_avg. The CPU is already part of nohz.idle_cpus_mask so the clear
	 * of nohz.has_blocked can only happen after checking the new load
	 */
9543
	if (rq->nohz_tick_stopped)
9544
		goto out;
9545

9546
	/* If we're a completely isolated CPU, we don't play: */
9547
	if (on_null_domain(rq))
9548 9549
		return;

9550 9551
	rq->nohz_tick_stopped = 1;

9552 9553
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
9554

9555 9556 9557 9558 9559 9560 9561
	/*
	 * Ensures that if nohz_idle_balance() fails to observe our
	 * @idle_cpus_mask store, it must observe the @has_blocked
	 * store.
	 */
	smp_mb__after_atomic();

9562
	set_cpu_sd_state_idle(cpu);
9563 9564 9565 9566 9567 9568 9569

out:
	/*
	 * Each time a cpu enter idle, we assume that it has blocked load and
	 * enable the periodic update of the load of idle cpus
	 */
	WRITE_ONCE(nohz.has_blocked, 1);
9570 9571 9572
}

/*
9573 9574 9575 9576 9577
 * Internal function that runs load balance for all idle cpus. The load balance
 * can be a simple update of blocked load or a complete load balance with
 * tasks movement depending of flags.
 * The function returns false if the loop has stopped before running
 * through all idle CPUs.
9578
 */
9579 9580
static bool _nohz_idle_balance(struct rq *this_rq, unsigned int flags,
			       enum cpu_idle_type idle)
9581
{
9582
	/* Earliest time when we have to do rebalance again */
9583 9584
	unsigned long now = jiffies;
	unsigned long next_balance = now + 60*HZ;
9585
	bool has_blocked_load = false;
9586
	int update_next_balance = 0;
P
Peter Zijlstra 已提交
9587 9588
	int this_cpu = this_rq->cpu;
	int balance_cpu;
9589
	int ret = false;
P
Peter Zijlstra 已提交
9590
	struct rq *rq;
9591

P
Peter Zijlstra 已提交
9592
	SCHED_WARN_ON((flags & NOHZ_KICK_MASK) == NOHZ_BALANCE_KICK);
9593

9594 9595 9596 9597 9598 9599 9600 9601 9602 9603 9604 9605 9606 9607 9608 9609
	/*
	 * We assume there will be no idle load after this update and clear
	 * the has_blocked flag. If a cpu enters idle in the mean time, it will
	 * set the has_blocked flag and trig another update of idle load.
	 * Because a cpu that becomes idle, is added to idle_cpus_mask before
	 * setting the flag, we are sure to not clear the state and not
	 * check the load of an idle cpu.
	 */
	WRITE_ONCE(nohz.has_blocked, 0);

	/*
	 * Ensures that if we miss the CPU, we must see the has_blocked
	 * store from nohz_balance_enter_idle().
	 */
	smp_mb();

9610
	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
9611
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
9612 9613 9614
			continue;

		/*
9615 9616
		 * If this CPU gets work to do, stop the load balancing
		 * work being done for other CPUs. Next load
9617 9618
		 * balancing owner will pick it up.
		 */
9619 9620 9621 9622
		if (need_resched()) {
			has_blocked_load = true;
			goto abort;
		}
9623

V
Vincent Guittot 已提交
9624 9625
		rq = cpu_rq(balance_cpu);

9626
		has_blocked_load |= update_nohz_stats(rq, true);
9627

9628 9629 9630 9631 9632
		/*
		 * If time for next balance is due,
		 * do the balance.
		 */
		if (time_after_eq(jiffies, rq->next_balance)) {
9633 9634
			struct rq_flags rf;

9635
			rq_lock_irqsave(rq, &rf);
9636
			update_rq_clock(rq);
9637
			cpu_load_update_idle(rq);
9638
			rq_unlock_irqrestore(rq, &rf);
9639

P
Peter Zijlstra 已提交
9640 9641
			if (flags & NOHZ_BALANCE_KICK)
				rebalance_domains(rq, CPU_IDLE);
9642
		}
9643

9644 9645 9646 9647
		if (time_after(next_balance, rq->next_balance)) {
			next_balance = rq->next_balance;
			update_next_balance = 1;
		}
9648
	}
9649

9650 9651 9652 9653 9654 9655
	/* Newly idle CPU doesn't need an update */
	if (idle != CPU_NEWLY_IDLE) {
		update_blocked_averages(this_cpu);
		has_blocked_load |= this_rq->has_blocked_load;
	}

P
Peter Zijlstra 已提交
9656 9657 9658
	if (flags & NOHZ_BALANCE_KICK)
		rebalance_domains(this_rq, CPU_IDLE);

9659 9660 9661
	WRITE_ONCE(nohz.next_blocked,
		now + msecs_to_jiffies(LOAD_AVG_PERIOD));

9662 9663 9664
	/* The full idle balance loop has been done */
	ret = true;

9665 9666 9667 9668
abort:
	/* There is still blocked load, enable periodic update */
	if (has_blocked_load)
		WRITE_ONCE(nohz.has_blocked, 1);
9669

9670 9671 9672 9673 9674 9675 9676
	/*
	 * next_balance will be updated only when there is a need.
	 * When the CPU is attached to null domain for ex, it will not be
	 * updated.
	 */
	if (likely(update_next_balance))
		nohz.next_balance = next_balance;
P
Peter Zijlstra 已提交
9677

9678 9679 9680 9681 9682 9683 9684 9685 9686 9687 9688 9689 9690 9691 9692 9693 9694 9695 9696 9697 9698 9699 9700 9701 9702 9703 9704 9705 9706
	return ret;
}

/*
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
static bool nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
{
	int this_cpu = this_rq->cpu;
	unsigned int flags;

	if (!(atomic_read(nohz_flags(this_cpu)) & NOHZ_KICK_MASK))
		return false;

	if (idle != CPU_IDLE) {
		atomic_andnot(NOHZ_KICK_MASK, nohz_flags(this_cpu));
		return false;
	}

	/*
	 * barrier, pairs with nohz_balance_enter_idle(), ensures ...
	 */
	flags = atomic_fetch_andnot(NOHZ_KICK_MASK, nohz_flags(this_cpu));
	if (!(flags & NOHZ_KICK_MASK))
		return false;

	_nohz_idle_balance(this_rq, flags, idle);

P
Peter Zijlstra 已提交
9707
	return true;
9708
}
9709 9710 9711 9712 9713 9714 9715 9716 9717 9718 9719 9720 9721 9722 9723 9724 9725 9726 9727 9728 9729 9730 9731 9732 9733 9734 9735 9736 9737 9738 9739 9740 9741

static void nohz_newidle_balance(struct rq *this_rq)
{
	int this_cpu = this_rq->cpu;

	/*
	 * This CPU doesn't want to be disturbed by scheduler
	 * housekeeping
	 */
	if (!housekeeping_cpu(this_cpu, HK_FLAG_SCHED))
		return;

	/* Will wake up very soon. No time for doing anything else*/
	if (this_rq->avg_idle < sysctl_sched_migration_cost)
		return;

	/* Don't need to update blocked load of idle CPUs*/
	if (!READ_ONCE(nohz.has_blocked) ||
	    time_before(jiffies, READ_ONCE(nohz.next_blocked)))
		return;

	raw_spin_unlock(&this_rq->lock);
	/*
	 * This CPU is going to be idle and blocked load of idle CPUs
	 * need to be updated. Run the ilb locally as it is a good
	 * candidate for ilb instead of waking up another idle CPU.
	 * Kick an normal ilb if we failed to do the update.
	 */
	if (!_nohz_idle_balance(this_rq, NOHZ_STATS_KICK, CPU_NEWLY_IDLE))
		kick_ilb(NOHZ_STATS_KICK);
	raw_spin_lock(&this_rq->lock);
}

9742 9743 9744
#else /* !CONFIG_NO_HZ_COMMON */
static inline void nohz_balancer_kick(struct rq *rq) { }

9745
static inline bool nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
P
Peter Zijlstra 已提交
9746 9747 9748
{
	return false;
}
9749 9750

static inline void nohz_newidle_balance(struct rq *this_rq) { }
9751
#endif /* CONFIG_NO_HZ_COMMON */
9752

P
Peter Zijlstra 已提交
9753 9754 9755 9756 9757 9758 9759 9760 9761 9762 9763 9764 9765 9766 9767 9768 9769 9770 9771 9772 9773 9774 9775 9776 9777 9778 9779 9780 9781 9782 9783 9784 9785 9786
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
static int idle_balance(struct rq *this_rq, struct rq_flags *rf)
{
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
	struct sched_domain *sd;
	int pulled_task = 0;
	u64 curr_cost = 0;

	/*
	 * We must set idle_stamp _before_ calling idle_balance(), such that we
	 * measure the duration of idle_balance() as idle time.
	 */
	this_rq->idle_stamp = rq_clock(this_rq);

	/*
	 * Do not pull tasks towards !active CPUs...
	 */
	if (!cpu_active(this_cpu))
		return 0;

	/*
	 * This is OK, because current is on_cpu, which avoids it being picked
	 * for load-balance and preemption/IRQs are still disabled avoiding
	 * further scheduler activity on it and we're being very careful to
	 * re-start the picking loop.
	 */
	rq_unpin_lock(this_rq, rf);

	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
9787

P
Peter Zijlstra 已提交
9788 9789 9790 9791 9792 9793
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
			update_next_balance(sd, &next_balance);
		rcu_read_unlock();

9794 9795
		nohz_newidle_balance(this_rq);

P
Peter Zijlstra 已提交
9796 9797 9798 9799 9800 9801 9802 9803 9804 9805 9806 9807 9808 9809 9810 9811 9812 9813 9814 9815 9816 9817 9818 9819 9820 9821 9822 9823 9824 9825 9826 9827 9828 9829 9830 9831 9832 9833 9834 9835 9836 9837 9838 9839 9840 9841 9842 9843 9844
		goto out;
	}

	raw_spin_unlock(&this_rq->lock);

	update_blocked_averages(this_cpu);
	rcu_read_lock();
	for_each_domain(this_cpu, sd) {
		int continue_balancing = 1;
		u64 t0, domain_cost;

		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
			update_next_balance(sd, &next_balance);
			break;
		}

		if (sd->flags & SD_BALANCE_NEWIDLE) {
			t0 = sched_clock_cpu(this_cpu);

			pulled_task = load_balance(this_cpu, this_rq,
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);

			domain_cost = sched_clock_cpu(this_cpu) - t0;
			if (domain_cost > sd->max_newidle_lb_cost)
				sd->max_newidle_lb_cost = domain_cost;

			curr_cost += domain_cost;
		}

		update_next_balance(sd, &next_balance);

		/*
		 * Stop searching for tasks to pull if there are
		 * now runnable tasks on this rq.
		 */
		if (pulled_task || this_rq->nr_running > 0)
			break;
	}
	rcu_read_unlock();

	raw_spin_lock(&this_rq->lock);

	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

9845
out:
P
Peter Zijlstra 已提交
9846 9847 9848 9849 9850 9851 9852 9853 9854 9855 9856 9857 9858 9859 9860 9861 9862 9863 9864 9865 9866 9867 9868 9869
	/*
	 * While browsing the domains, we released the rq lock, a task could
	 * have been enqueued in the meantime. Since we're not going idle,
	 * pretend we pulled a task.
	 */
	if (this_rq->cfs.h_nr_running && !pulled_task)
		pulled_task = 1;

	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
		this_rq->next_balance = next_balance;

	/* Is there a task of a high priority class? */
	if (this_rq->nr_running != this_rq->cfs.h_nr_running)
		pulled_task = -1;

	if (pulled_task)
		this_rq->idle_stamp = 0;

	rq_repin_lock(this_rq, rf);

	return pulled_task;
}

9870 9871 9872 9873
/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
9874
static __latent_entropy void run_rebalance_domains(struct softirq_action *h)
9875
{
9876
	struct rq *this_rq = this_rq();
9877
	enum cpu_idle_type idle = this_rq->idle_balance ?
9878 9879 9880
						CPU_IDLE : CPU_NOT_IDLE;

	/*
9881 9882
	 * If this CPU has a pending nohz_balance_kick, then do the
	 * balancing on behalf of the other idle CPUs whose ticks are
9883
	 * stopped. Do nohz_idle_balance *before* rebalance_domains to
9884
	 * give the idle CPUs a chance to load balance. Else we may
9885 9886
	 * load balance only within the local sched_domain hierarchy
	 * and abort nohz_idle_balance altogether if we pull some load.
9887
	 */
P
Peter Zijlstra 已提交
9888 9889 9890 9891 9892
	if (nohz_idle_balance(this_rq, idle))
		return;

	/* normal load balance */
	update_blocked_averages(this_rq->cpu);
9893
	rebalance_domains(this_rq, idle);
9894 9895 9896 9897 9898
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
9899
void trigger_load_balance(struct rq *rq)
9900 9901
{
	/* Don't need to rebalance while attached to NULL domain */
9902 9903 9904 9905
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
9906
		raise_softirq(SCHED_SOFTIRQ);
9907 9908

	nohz_balancer_kick(rq);
9909 9910
}

9911 9912 9913
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
9914 9915

	update_runtime_enabled(rq);
9916 9917 9918 9919 9920
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
9921 9922 9923

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
9924 9925
}

9926
#endif /* CONFIG_SMP */
9927

9928
/*
9929 9930 9931 9932 9933 9934
 * scheduler tick hitting a task of our scheduling class.
 *
 * NOTE: This function can be called remotely by the tick offload that
 * goes along full dynticks. Therefore no local assumption can be made
 * and everything must be accessed through the @rq and @curr passed in
 * parameters.
9935
 */
P
Peter Zijlstra 已提交
9936
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
9937 9938 9939 9940 9941 9942
{
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &curr->se;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
P
Peter Zijlstra 已提交
9943
		entity_tick(cfs_rq, se, queued);
9944
	}
9945

9946
	if (static_branch_unlikely(&sched_numa_balancing))
9947
		task_tick_numa(rq, curr);
9948 9949 9950
}

/*
P
Peter Zijlstra 已提交
9951 9952 9953
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
9954
 */
P
Peter Zijlstra 已提交
9955
static void task_fork_fair(struct task_struct *p)
9956
{
9957 9958
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
P
Peter Zijlstra 已提交
9959
	struct rq *rq = this_rq();
9960
	struct rq_flags rf;
9961

9962
	rq_lock(rq, &rf);
9963 9964
	update_rq_clock(rq);

9965 9966
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;
9967 9968
	if (curr) {
		update_curr(cfs_rq);
9969
		se->vruntime = curr->vruntime;
9970
	}
9971
	place_entity(cfs_rq, se, 1);
9972

P
Peter Zijlstra 已提交
9973
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
9974
		/*
9975 9976 9977
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
9978
		swap(curr->vruntime, se->vruntime);
9979
		resched_curr(rq);
9980
	}
9981

9982
	se->vruntime -= cfs_rq->min_vruntime;
9983
	rq_unlock(rq, &rf);
9984 9985
}

9986 9987 9988 9989
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
9990 9991
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
9992
{
9993
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
9994 9995
		return;

9996 9997 9998 9999 10000
	/*
	 * Reschedule if we are currently running on this runqueue and
	 * our priority decreased, or if we are not currently running on
	 * this runqueue and our priority is higher than the current's
	 */
P
Peter Zijlstra 已提交
10001
	if (rq->curr == p) {
10002
		if (p->prio > oldprio)
10003
			resched_curr(rq);
10004
	} else
10005
		check_preempt_curr(rq, p, 0);
10006 10007
}

10008
static inline bool vruntime_normalized(struct task_struct *p)
P
Peter Zijlstra 已提交
10009 10010 10011 10012
{
	struct sched_entity *se = &p->se;

	/*
10013 10014 10015 10016 10017 10018 10019 10020 10021 10022
	 * In both the TASK_ON_RQ_QUEUED and TASK_ON_RQ_MIGRATING cases,
	 * the dequeue_entity(.flags=0) will already have normalized the
	 * vruntime.
	 */
	if (p->on_rq)
		return true;

	/*
	 * When !on_rq, vruntime of the task has usually NOT been normalized.
	 * But there are some cases where it has already been normalized:
P
Peter Zijlstra 已提交
10023
	 *
10024 10025 10026 10027
	 * - A forked child which is waiting for being woken up by
	 *   wake_up_new_task().
	 * - A task which has been woken up by try_to_wake_up() and
	 *   waiting for actually being woken up by sched_ttwu_pending().
P
Peter Zijlstra 已提交
10028
	 */
10029 10030 10031 10032 10033 10034
	if (!se->sum_exec_runtime || p->state == TASK_WAKING)
		return true;

	return false;
}

10035 10036 10037 10038 10039 10040 10041 10042 10043 10044 10045 10046 10047 10048 10049 10050 10051 10052
#ifdef CONFIG_FAIR_GROUP_SCHED
/*
 * Propagate the changes of the sched_entity across the tg tree to make it
 * visible to the root
 */
static void propagate_entity_cfs_rq(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq;

	/* Start to propagate at parent */
	se = se->parent;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);

		if (cfs_rq_throttled(cfs_rq))
			break;

10053
		update_load_avg(cfs_rq, se, UPDATE_TG);
10054 10055 10056 10057 10058 10059
	}
}
#else
static void propagate_entity_cfs_rq(struct sched_entity *se) { }
#endif

10060
static void detach_entity_cfs_rq(struct sched_entity *se)
10061 10062 10063
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

10064
	/* Catch up with the cfs_rq and remove our load when we leave */
10065
	update_load_avg(cfs_rq, se, 0);
10066
	detach_entity_load_avg(cfs_rq, se);
10067
	update_tg_load_avg(cfs_rq, false);
10068
	propagate_entity_cfs_rq(se);
P
Peter Zijlstra 已提交
10069 10070
}

10071
static void attach_entity_cfs_rq(struct sched_entity *se)
10072
{
10073
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
10074 10075

#ifdef CONFIG_FAIR_GROUP_SCHED
10076 10077 10078 10079 10080 10081
	/*
	 * Since the real-depth could have been changed (only FAIR
	 * class maintain depth value), reset depth properly.
	 */
	se->depth = se->parent ? se->parent->depth + 1 : 0;
#endif
10082

10083
	/* Synchronize entity with its cfs_rq */
10084
	update_load_avg(cfs_rq, se, sched_feat(ATTACH_AGE_LOAD) ? 0 : SKIP_AGE_LOAD);
10085
	attach_entity_load_avg(cfs_rq, se, 0);
10086
	update_tg_load_avg(cfs_rq, false);
10087
	propagate_entity_cfs_rq(se);
10088 10089 10090 10091 10092 10093 10094 10095 10096 10097 10098 10099 10100 10101 10102 10103 10104 10105 10106 10107 10108 10109 10110 10111 10112
}

static void detach_task_cfs_rq(struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	if (!vruntime_normalized(p)) {
		/*
		 * Fix up our vruntime so that the current sleep doesn't
		 * cause 'unlimited' sleep bonus.
		 */
		place_entity(cfs_rq, se, 0);
		se->vruntime -= cfs_rq->min_vruntime;
	}

	detach_entity_cfs_rq(se);
}

static void attach_task_cfs_rq(struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	attach_entity_cfs_rq(se);
10113 10114 10115 10116

	if (!vruntime_normalized(p))
		se->vruntime += cfs_rq->min_vruntime;
}
10117

10118 10119 10120 10121 10122 10123 10124 10125
static void switched_from_fair(struct rq *rq, struct task_struct *p)
{
	detach_task_cfs_rq(p);
}

static void switched_to_fair(struct rq *rq, struct task_struct *p)
{
	attach_task_cfs_rq(p);
10126

10127
	if (task_on_rq_queued(p)) {
10128
		/*
10129 10130 10131
		 * We were most likely switched from sched_rt, so
		 * kick off the schedule if running, otherwise just see
		 * if we can still preempt the current task.
10132
		 */
10133 10134 10135 10136
		if (rq->curr == p)
			resched_curr(rq);
		else
			check_preempt_curr(rq, p, 0);
10137
	}
10138 10139
}

10140 10141 10142 10143 10144 10145 10146 10147 10148
/* Account for a task changing its policy or group.
 *
 * This routine is mostly called to set cfs_rq->curr field when a task
 * migrates between groups/classes.
 */
static void set_curr_task_fair(struct rq *rq)
{
	struct sched_entity *se = &rq->curr->se;

10149 10150 10151 10152 10153 10154 10155
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);

		set_next_entity(cfs_rq, se);
		/* ensure bandwidth has been allocated on our new cfs_rq */
		account_cfs_rq_runtime(cfs_rq, 0);
	}
10156 10157
}

10158 10159
void init_cfs_rq(struct cfs_rq *cfs_rq)
{
10160
	cfs_rq->tasks_timeline = RB_ROOT_CACHED;
10161 10162 10163 10164
	cfs_rq->min_vruntime = (u64)(-(1LL << 20));
#ifndef CONFIG_64BIT
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
10165
#ifdef CONFIG_SMP
10166
	raw_spin_lock_init(&cfs_rq->removed.lock);
10167
#endif
10168 10169
}

P
Peter Zijlstra 已提交
10170
#ifdef CONFIG_FAIR_GROUP_SCHED
10171 10172 10173 10174 10175 10176 10177 10178
static void task_set_group_fair(struct task_struct *p)
{
	struct sched_entity *se = &p->se;

	set_task_rq(p, task_cpu(p));
	se->depth = se->parent ? se->parent->depth + 1 : 0;
}

10179
static void task_move_group_fair(struct task_struct *p)
P
Peter Zijlstra 已提交
10180
{
10181
	detach_task_cfs_rq(p);
10182
	set_task_rq(p, task_cpu(p));
10183 10184 10185 10186 10187

#ifdef CONFIG_SMP
	/* Tell se's cfs_rq has been changed -- migrated */
	p->se.avg.last_update_time = 0;
#endif
10188
	attach_task_cfs_rq(p);
P
Peter Zijlstra 已提交
10189
}
10190

10191 10192 10193 10194 10195 10196 10197 10198 10199 10200 10201 10202 10203
static void task_change_group_fair(struct task_struct *p, int type)
{
	switch (type) {
	case TASK_SET_GROUP:
		task_set_group_fair(p);
		break;

	case TASK_MOVE_GROUP:
		task_move_group_fair(p);
		break;
	}
}

10204 10205 10206 10207 10208 10209 10210 10211 10212
void free_fair_sched_group(struct task_group *tg)
{
	int i;

	destroy_cfs_bandwidth(tg_cfs_bandwidth(tg));

	for_each_possible_cpu(i) {
		if (tg->cfs_rq)
			kfree(tg->cfs_rq[i]);
10213
		if (tg->se)
10214 10215 10216 10217 10218 10219 10220 10221 10222 10223
			kfree(tg->se[i]);
	}

	kfree(tg->cfs_rq);
	kfree(tg->se);
}

int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
{
	struct sched_entity *se;
10224
	struct cfs_rq *cfs_rq;
10225 10226 10227 10228 10229 10230 10231 10232 10233 10234 10235 10236 10237 10238 10239 10240 10241 10242 10243 10244 10245 10246 10247 10248 10249 10250
	int i;

	tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
	if (!tg->cfs_rq)
		goto err;
	tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
	if (!tg->se)
		goto err;

	tg->shares = NICE_0_LOAD;

	init_cfs_bandwidth(tg_cfs_bandwidth(tg));

	for_each_possible_cpu(i) {
		cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
				      GFP_KERNEL, cpu_to_node(i));
		if (!cfs_rq)
			goto err;

		se = kzalloc_node(sizeof(struct sched_entity),
				  GFP_KERNEL, cpu_to_node(i));
		if (!se)
			goto err_free_rq;

		init_cfs_rq(cfs_rq);
		init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]);
10251
		init_entity_runnable_average(se);
10252 10253 10254 10255 10256 10257 10258 10259 10260 10261
	}

	return 1;

err_free_rq:
	kfree(cfs_rq);
err:
	return 0;
}

10262 10263 10264 10265 10266 10267 10268 10269 10270 10271 10272
void online_fair_sched_group(struct task_group *tg)
{
	struct sched_entity *se;
	struct rq *rq;
	int i;

	for_each_possible_cpu(i) {
		rq = cpu_rq(i);
		se = tg->se[i];

		raw_spin_lock_irq(&rq->lock);
10273
		update_rq_clock(rq);
10274
		attach_entity_cfs_rq(se);
10275
		sync_throttle(tg, i);
10276 10277 10278 10279
		raw_spin_unlock_irq(&rq->lock);
	}
}

10280
void unregister_fair_sched_group(struct task_group *tg)
10281 10282
{
	unsigned long flags;
10283 10284
	struct rq *rq;
	int cpu;
10285

10286 10287 10288
	for_each_possible_cpu(cpu) {
		if (tg->se[cpu])
			remove_entity_load_avg(tg->se[cpu]);
10289

10290 10291 10292 10293 10294 10295 10296 10297 10298 10299 10300 10301 10302
		/*
		 * Only empty task groups can be destroyed; so we can speculatively
		 * check on_list without danger of it being re-added.
		 */
		if (!tg->cfs_rq[cpu]->on_list)
			continue;

		rq = cpu_rq(cpu);

		raw_spin_lock_irqsave(&rq->lock, flags);
		list_del_leaf_cfs_rq(tg->cfs_rq[cpu]);
		raw_spin_unlock_irqrestore(&rq->lock, flags);
	}
10303 10304 10305 10306 10307 10308 10309 10310 10311 10312 10313 10314 10315 10316 10317 10318 10319 10320 10321
}

void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
			struct sched_entity *se, int cpu,
			struct sched_entity *parent)
{
	struct rq *rq = cpu_rq(cpu);

	cfs_rq->tg = tg;
	cfs_rq->rq = rq;
	init_cfs_rq_runtime(cfs_rq);

	tg->cfs_rq[cpu] = cfs_rq;
	tg->se[cpu] = se;

	/* se could be NULL for root_task_group */
	if (!se)
		return;

P
Peter Zijlstra 已提交
10322
	if (!parent) {
10323
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
10324 10325
		se->depth = 0;
	} else {
10326
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
10327 10328
		se->depth = parent->depth + 1;
	}
10329 10330

	se->my_q = cfs_rq;
10331 10332
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
10333 10334 10335 10336 10337 10338 10339 10340 10341 10342 10343 10344 10345 10346 10347 10348 10349 10350 10351 10352 10353 10354 10355 10356
	se->parent = parent;
}

static DEFINE_MUTEX(shares_mutex);

int sched_group_set_shares(struct task_group *tg, unsigned long shares)
{
	int i;

	/*
	 * We can't change the weight of the root cgroup.
	 */
	if (!tg->se[0])
		return -EINVAL;

	shares = clamp(shares, scale_load(MIN_SHARES), scale_load(MAX_SHARES));

	mutex_lock(&shares_mutex);
	if (tg->shares == shares)
		goto done;

	tg->shares = shares;
	for_each_possible_cpu(i) {
		struct rq *rq = cpu_rq(i);
10357 10358
		struct sched_entity *se = tg->se[i];
		struct rq_flags rf;
10359 10360

		/* Propagate contribution to hierarchy */
10361
		rq_lock_irqsave(rq, &rf);
10362
		update_rq_clock(rq);
10363
		for_each_sched_entity(se) {
10364
			update_load_avg(cfs_rq_of(se), se, UPDATE_TG);
10365
			update_cfs_group(se);
10366
		}
10367
		rq_unlock_irqrestore(rq, &rf);
10368 10369 10370 10371 10372 10373 10374 10375 10376 10377 10378 10379 10380 10381 10382
	}

done:
	mutex_unlock(&shares_mutex);
	return 0;
}
#else /* CONFIG_FAIR_GROUP_SCHED */

void free_fair_sched_group(struct task_group *tg) { }

int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
{
	return 1;
}

10383 10384
void online_fair_sched_group(struct task_group *tg) { }

10385
void unregister_fair_sched_group(struct task_group *tg) { }
10386 10387 10388

#endif /* CONFIG_FAIR_GROUP_SCHED */

P
Peter Zijlstra 已提交
10389

10390
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
10391 10392 10393 10394 10395 10396 10397 10398 10399
{
	struct sched_entity *se = &task->se;
	unsigned int rr_interval = 0;

	/*
	 * Time slice is 0 for SCHED_OTHER tasks that are on an otherwise
	 * idle runqueue:
	 */
	if (rq->cfs.load.weight)
10400
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
10401 10402 10403 10404

	return rr_interval;
}

10405 10406 10407
/*
 * All the scheduling class methods:
 */
10408
const struct sched_class fair_sched_class = {
10409
	.next			= &idle_sched_class,
10410 10411 10412
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
10413
	.yield_to_task		= yield_to_task_fair,
10414

I
Ingo Molnar 已提交
10415
	.check_preempt_curr	= check_preempt_wakeup,
10416 10417 10418 10419

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

10420
#ifdef CONFIG_SMP
L
Li Zefan 已提交
10421
	.select_task_rq		= select_task_rq_fair,
10422
	.migrate_task_rq	= migrate_task_rq_fair,
10423

10424 10425
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
10426

10427
	.task_dead		= task_dead_fair,
10428
	.set_cpus_allowed	= set_cpus_allowed_common,
10429
#endif
10430

10431
	.set_curr_task          = set_curr_task_fair,
10432
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
10433
	.task_fork		= task_fork_fair,
10434 10435

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
10436
	.switched_from		= switched_from_fair,
10437
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
10438

10439 10440
	.get_rr_interval	= get_rr_interval_fair,

10441 10442
	.update_curr		= update_curr_fair,

P
Peter Zijlstra 已提交
10443
#ifdef CONFIG_FAIR_GROUP_SCHED
10444
	.task_change_group	= task_change_group_fair,
P
Peter Zijlstra 已提交
10445
#endif
10446 10447 10448
};

#ifdef CONFIG_SCHED_DEBUG
10449
void print_cfs_stats(struct seq_file *m, int cpu)
10450
{
10451
	struct cfs_rq *cfs_rq, *pos;
10452

10453
	rcu_read_lock();
10454
	for_each_leaf_cfs_rq_safe(cpu_rq(cpu), cfs_rq, pos)
10455
		print_cfs_rq(m, cpu, cfs_rq);
10456
	rcu_read_unlock();
10457
}
10458 10459 10460 10461 10462 10463 10464 10465 10466 10467 10468 10469 10470 10471 10472 10473 10474 10475 10476 10477 10478

#ifdef CONFIG_NUMA_BALANCING
void show_numa_stats(struct task_struct *p, struct seq_file *m)
{
	int node;
	unsigned long tsf = 0, tpf = 0, gsf = 0, gpf = 0;

	for_each_online_node(node) {
		if (p->numa_faults) {
			tsf = p->numa_faults[task_faults_idx(NUMA_MEM, node, 0)];
			tpf = p->numa_faults[task_faults_idx(NUMA_MEM, node, 1)];
		}
		if (p->numa_group) {
			gsf = p->numa_group->faults[task_faults_idx(NUMA_MEM, node, 0)],
			gpf = p->numa_group->faults[task_faults_idx(NUMA_MEM, node, 1)];
		}
		print_numa_stats(m, node, tsf, tpf, gsf, gpf);
	}
}
#endif /* CONFIG_NUMA_BALANCING */
#endif /* CONFIG_SCHED_DEBUG */
10479 10480 10481 10482 10483 10484

__init void init_sched_fair_class(void)
{
#ifdef CONFIG_SMP
	open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);

10485
#ifdef CONFIG_NO_HZ_COMMON
10486
	nohz.next_balance = jiffies;
10487
	nohz.next_blocked = jiffies;
10488 10489 10490 10491 10492
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
#endif
#endif /* SMP */

}